CN1093966C - Apparatus and method for character identification - Google Patents

Abstract

The purpose is to recognize characters with high precision and high speed. Input handwritten character (input figure) by input unit 1, extract the stroke feature that forms input handwritten character by stroke extracting unit 2, by stroke classifying unit 4, stroke is classified into linear stroke, non-linear stroke and successor character stroke according to this feature, will be by Straight-line strokes correspond to connection unit 5, non-linear strokes correspond to connection unit 7 and successor word strokes correspond to connection unit 8 to classify the feature of each stroke and the stroke feature of the recognition object character (standard figure) stored in the stroke feature dictionary 3 to connect correspondingly, Handwritten characters are recognized by the character evaluation unit 10 based on the result.

Description

Character recognition device and character recognition method

The present invention relates to reading handwritten characters written on an input board with a pen and performing character recognition on the handwritten characters. It relates to a character recognition device and a character recognition method. The characteristics of strokes are used to identify the recognized object text.

The prior online handwritten character recognition device that absorbs both the change of stroke order and the change of the number of strokes describes the strokes of handwritten characters with feature point strings. The complementary optimal corresponding connection algorithm absorbs stroke order changes by making one-to-one correspondence between the strokes of handwritten characters and graphics including changes in stroke order and number of strokes and the strokes of standard characters and graphics represented by standard scripts with stroke order and number of strokes. , use the corresponding connection with the smallest distance when the strokes are combined to absorb the change in the number of strokes, and perform character recognition (online handwritten Chinese character recognition by stroke-corresponding connections, NTT R&D Vol. 45 No. 11, 1996).

For example, FIG. 34 is a block diagram showing the structure of a conventional device for absorbing changes in the order of strokes and the number of strokes shown in the prior art. Among Fig. 34, 201 is the input unit that inputs the handwritten characters written on the input board with a pen, i.e. the input graphics, 202 is a preprocessing unit that samples the input graphics and normalizes the position size, etc., and 203 is the input graphics that will be formed Each stroke is engraved at equal intervals as a feature point extraction unit for feature point extraction. 204 is a standard graphic dictionary that has stored the average character graphics that are made by a plurality of italic characters written with standard stroke order and number of strokes, that is, standard graphics. 205 It is an inter-stroke distance calculation unit for calculating the distance between the strokes of the input graphics and the strokes of the standard graphics. 206 uses two complementary search algorithms to connect the strokes of the input graphics and the strokes of the standard graphics in one-to-one correspondence to absorb stroke order changes Stroke-to-one correspondence connection unit, 207 is the selection stroke combination unit that absorbs the change in the number of strokes by combining uncorresponding strokes in the standard graphics with the front and rear strokes, 208 is the inter-graphic distance calculation unit that calculates the distance between graphics according to the corresponding connection results, 209 is a control unit for controlling each unit.

The operation of the conventional example will be described using the block diagram of Fig. 34 . First, the control unit 209 instructs the input unit 201 to obtain an input pattern.

Next, the control unit 209 sends the input graphics input by the input unit 201 to the pre-processing unit 202 for sampling and standardization of position and size.

The control unit 209 sends the pre-processed input graphic to the feature point extraction unit 203, and converts the input graphic into a string of feature points engraved on each stroke of the input graphic at equal intervals.

The control unit 209 sends the feature point string of the input figure to the one-to-one stroke connection unit 206 . The one-to-one correspondence stroke connection unit 206 is based on the stroke of the side with the few strokes as a reference, and the input graphics and the standard graphics of the standard graphics dictionary 204 are connected correspondingly, so that each feature point does not have redundant corresponding connections (corresponding connections of the excess corresponding cancellation type) algorithm).

Then, the one-to-one correspondence stroke connection unit 206 is based on the stroke of the side with the largest number of strokes, and the input graphics and the standard graphics of the standard graphics dictionary 204 are connected correspondingly, so that each feature point does not have insufficient corresponding connections (insufficient corresponding release type) corresponding to the join algorithm).

Among the corresponding connection results obtained by the above two corresponding connection algorithms, the result with the smaller distance is taken as the final result obtained by the one-to-one corresponding stroke connection unit 206 .

Here, among the distances calculated by the stroke-to-stroke distance calculating section 205 at the time of corresponding connection, the value obtained by dividing the sum of the distances between the start points and the end points, that is, the end point matching distance, and the point of the stroke with the smaller number of points are used. Partial matching is performed by sequentially correspondingly connecting the first point of the stroke with the larger number of points, and multiplying the sum of the distance between the points by the point ratio.

Next, the selected stroke combining unit 207 matches the order of the strokes connected by the one-to-one corresponding stroke connecting unit 206 with the stroke order of the side with more strokes, and changes the line. For the strokes that are not connected by the one-to-one corresponding stroke connection unit 206, stroke combination is performed based on the assumption of "generating stroke combination according to the stroke order".

Specifically, for uncorresponding strokes written earlier than the first strokes of the strokes correspondingly connected by the one-to-one corresponding stroke connecting unit 206, they are combined with the first strokes in stroke order. Similarly, the uncorresponding strokes written after comparing with the final strokes are combined with the final strokes according to the order of the strokes. Other non-corresponding strokes exist between arbitrary strokes of the strokes correspondingly connected by the one-to-one corresponding stroke connecting unit 206. Therefore, the uncorresponding strokes are temporarily combined with the preceding and following strokes, and the distance between the two strokes is minimized. Open to split into two strokes.

Here, in the selected stroke combination unit 207, remove the dots of the stroke with more dots at equal intervals, make it consistent with the dots of the less dot side, and then connect correspondingly, calculate the sum of the distances between points, and use the dot with a few dots. The value divided by the number of points on one side is the overall matching distance. However, for high-speed processing, the endpoint matching distance is used in large classification processing.

Finally, according to the result of the corresponding connection, appropriate distance normalization is performed on the combined strokes, the final distance is calculated, and the character with the smallest distance is taken as the recognition result.

In the above-described character recognition device that absorbs changes in the order of strokes and the number of strokes in the conventional example, the distance between strokes is calculated using a string of feature points carved on each stroke of a handwritten character at equal intervals, so there are noise components such as left and right strokes. There is a problem that the stroke distance is affected by the offset of the coordinate value of the feature point due to the stroke deformation and the position offset, which causes misreading.

Also, in characters with many strokes such as Chinese characters, each stroke is mostly composed of simple straight strokes, so it does not matter even if the feature points in the strokes do not include important information, because the corresponding connection is performed with the feature point values, therefore, in There is a problem that it takes time to connect and is easily affected by coordinate value offset.

In addition, when absorbing changes in the number of strokes, in order to reduce the amount of calculation, the assumption of "generating stroke combinations according to the stroke order" is used. When the change in the number of strokes does not produce changes according to the stroke order, there is a problem that it cannot respond.

The present invention was made to solve the above problems, and its object is to provide such a character recognition device and character recognition method, that is, the strokes of handwritten characters that can completely change the stroke order and strokes are connected in correspondence with the strokes of the stroke feature dictionary, and it is difficult to be affected. Due to the influence of local deformation and positional deviation of strokes, high-precision recognition can be realized, and high-speed character recognition can also be performed.

Regarding the first aspect of the present invention, it includes: an input unit for online input of handwritten characters; a stroke feature dictionary for pre-storing the features of the linear strokes and non-linear strokes that constitute each recognition target character for a plurality of recognition target characters; a stroke feature extraction unit extracting the features of the strokes that constitute the handwritten characters from the above-mentioned handwritten characters input by the above-mentioned input unit; the stroke classification unit classifies each stroke of the above-mentioned handwritten characters into the above-mentioned linear strokes, the above-mentioned non-linear strokes according to the features extracted by the stroke feature extraction unit One of the strokes of straight lines or continuous characters that cannot be correspondingly connected with the strokes that constitute the above-mentioned recognition object characters; the straight-line strokes correspond to the connection unit, and the features of the straight-line strokes classified by the stroke classification unit are identified with the composition stored in the above-mentioned stroke feature dictionary The corresponding connection of the features of the linear strokes of the object characters; the non-linear strokes correspond to the connection unit, and carry out the feature of the non-linear strokes classified by the stroke classification unit and the features of the non-linear strokes that constitute the recognition object characters stored in the above-mentioned stroke feature dictionary Correspondingly connected; the corresponding connection unit of the subsequent character stroke, the subsequent character stroke classified by the above-mentioned stroke classification unit is divided into the above-mentioned linear stroke or the above-mentioned non-linear stroke, and the feature of the linear stroke or non-linear stroke of the segmentation is stored in the above-mentioned stroke feature dictionary The corresponding connection of the features of the linear strokes or non-linear strokes that constitute the recognition object text; the character evaluation unit is based on the corresponding connection obtained by the corresponding connection unit of the above-mentioned straight strokes, the corresponding connection unit of the above-mentioned non-linear strokes or the corresponding connection unit of the above-mentioned subsequent character strokes As a result, the above handwritten characters were recognized.

Regarding the second aspect of the present invention, it also includes a stroke corresponding connection determination unit, which determines the existence area of each stroke of the above-mentioned handwritten character according to the features extracted by the above-mentioned stroke feature extraction unit; the above-mentioned linear stroke corresponding connection unit, the above-mentioned non-linear connection stroke The corresponding connection unit or the above-mentioned successor character stroke corresponding connection unit performs the linear strokes or non-linear strokes corresponding to each stroke existence area determined by the above-mentioned corresponding connection area determination unit and constituting the recognition object character stored in the above-mentioned stroke feature dictionary. Corresponding connections of features.

According to the third aspect of the present invention, the above-mentioned non-linear stroke corresponding connection unit performs corresponding connection on the above-mentioned recognition characters correspondingly connected by the above-mentioned linear stroke corresponding connection unit.

Regarding the fourth aspect of the present invention, the subsequent character stroke corresponding connection unit performs corresponding connection on the above-mentioned recognition object characters correspondingly connected by the above-mentioned linear stroke corresponding connection unit or the above-mentioned non-linear stroke corresponding connection unit.

Regarding the fifth aspect of the present invention, it includes: an input step of inputting handwritten characters online; a stroke feature extraction step of extracting features of strokes constituting the handwritten characters from the above-mentioned handwritten characters input by the above input step; according to the stroke feature extraction step The feature of extracting, each stroke of above-mentioned handwritten character is classified into one of above-mentioned linear stroke, above-mentioned non-linear stroke and the stroke that can not be correspondingly connected with the stroke that forms above-mentioned recognition object character; Carry out by this stroke classification The linear stroke feature of the step classification and the corresponding connection step of the linear stroke that constitutes the feature of the linear stroke of the recognition object character stored in the above-mentioned stroke feature dictionary; The non-linear stroke corresponding connection step of the corresponding connection of the feature of the non-linear stroke of the recognition object character stored in the feature dictionary; the subsequent character strokes classified by the above-mentioned stroke classification step are divided into the above-mentioned linear stroke or the above-mentioned non-linear stroke, and the segmentation is carried out. The features of the linear strokes or non-linear strokes and the features of the linear strokes or non-linear strokes that constitute the recognition object characters stored in the above-mentioned stroke feature dictionary correspond to the corresponding connection steps of the subsequent character strokes; and according to the above-mentioned linear strokes. A character evaluation step for recognizing the above-mentioned handwritten characters by the corresponding connection result obtained in the non-linear stroke corresponding connection step or the above-mentioned subsequent character stroke corresponding connection step.

Regarding the sixth aspect of the present invention, it includes determining the stroke-corresponding connecting region of the existence region of each stroke of the above-mentioned handwritten characters according to the features extracted by the above-mentioned stroke feature extraction step;

The above-mentioned linear stroke connection step, the above-mentioned non-linear stroke connection step or the corresponding connection step of the above-mentioned subsequent character strokes are carried out corresponding to the existence area of each stroke determined in the above-mentioned corresponding connection area determination step, and the composition is stored in the corresponding existence area of each stroke. Corresponding connection of features of straight strokes or non-straight strokes of the character to be recognized in the above-mentioned stroke feature dictionary.

According to the seventh aspect of the present invention, the non-linear stroke corresponding connection step performs corresponding connection on the above-mentioned recognition target characters correspondingly connected by the above-mentioned linear stroke corresponding step.

Regarding the eighth aspect of the present invention, the subsequent character stroke corresponding connection step performs corresponding connection on the above-mentioned recognition object characters correspondingly connected by the above-mentioned linear stroke corresponding step or the above-mentioned non-linear stroke corresponding connection step.

FIG. 1 is a block diagram of a character recognition device according to the first embodiment.

FIG. 2 is a flowchart showing the processing flow of the character recognition device of the first embodiment.

FIG. 3 is a schematic diagram showing the input pattern "sub" written in the input unit 1 of the first embodiment.

FIG. 4 is a schematic diagram showing the stroke features of the input figure extracted by the stroke feature extraction unit 2 in the first embodiment.

FIG. 5 is an allocation diagram showing stroke directions and virtual stroke directions quantized into 16 directions in the stroke feature of the first embodiment.

FIG. 6 is a flowchart showing the processing flow of the stroke classifying unit in the first embodiment.

Fig. 7 is a diagram showing the distribution of eight direction components used when the stroke classifying unit in the first embodiment obtains the distribution of direction components.

FIG. 8 is a flowchart showing the processing flow of the straight-line-stroke-corresponding connection device 5 according to the first embodiment.

FIG. 9 is a diagram showing the division of regions used by the stroke-corresponding-connection region determining unit 6 of the first embodiment when connecting straight-line strokes.

FIG. 10 is a diagram showing information for determining a corresponding connection region used by the stroke-corresponding connection region determining unit 6 of the first embodiment.

FIG. 11 is a diagram showing the connection area to which an input stroke can be corresponded in the first embodiment.

Fig. 12 is a stroke string diagram showing a standard pattern of the character "亚" in the stroke feature dictionary of the first embodiment.

Fig. 13 is a graph showing stroke characteristics in the standard pattern of the character "亚" in the stroke characteristic dictionary of the first embodiment.

FIG. 14 is a diagram showing whether a corresponding connection is possible according to stroke shapes used in the linear stroke corresponding connection unit 5 , the non-linear stroke corresponding connection unit 7 , and the subsequent character stroke corresponding connection unit 8 according to the first embodiment.

FIG. 15 is a view showing a list of stroke candidates in the linear stroke-corresponding connection unit 5 of the first embodiment.

FIG. 16 is a flowchart showing the processing flow of the non-linear stroke correspondence connection unit 7 in the first embodiment.

FIG. 17 is a flow chart showing the flow of processing by the successor-character-stroke-correspondence-connection unit 8 of the first embodiment.

FIG. 18 is a diagram showing the division of regions used by the stroke-corresponding-connection-area determination device 6 of Embodiment 1 in the case of corresponding connections other than straight strokes.

FIG. 19 is a diagram showing in which area each stroke of the input pattern according to the first embodiment exists in the stroke-corresponding connectable area determining means 6 .

FIG. 20 is a flowchart showing the detailed processing flow of step S54 in the first embodiment.

FIG. 21 is a diagram showing feature points of strokes of standard figures in the first embodiment.

Fig. 22 is a diagram showing sampling points for sampling the strokes of the input pattern in the first embodiment at regular intervals.

FIG. 23 is a diagram showing the result of correspondence connection between the feature points of the standard graph and the sampling points of the input graph in the first embodiment.

Fig. 24 is a diagram showing strokes decomposed by making the input figure of the first embodiment coincide with the strokes of the standard figure.

FIG. 25 is a diagram showing individual strokes of the character "言" written by the input unit 1 of the first embodiment.

FIG. 26 is a diagram showing in which area each stroke of the input pattern according to the first embodiment exists in the stroke-corresponding connectable area determination unit 6 .

Fig. 27 is a diagram showing standard patterns of the character "yan" in the stroke feature dictionary 3 of the first embodiment.

Fig. 28 is a flowchart showing the detailed processing flow of the partial stroke correspondence connection in the first embodiment.

Fig. 29 is a diagram showing decomposition candidate points of subsequent character strokes of an input pattern according to the first embodiment.

FIG. 30 is a diagram showing partial strokes in the first embodiment.

FIG. 31 is a flowchart showing the detailed flow of the processing of step S72 in FIG. 28 .

FIG. 32 is a diagram showing decomposed strokes obtained from partial strokes in FIG. 30 .

Fig. 33 is a diagram showing decomposed strokes obtained from partial strokes.

Fig. 34 is a structural diagram of a conventional online character recognition device.

(Example 1)

Next, Embodiment 1 will be described with reference to the drawings.

FIG. 1 is a block diagram of a character recognition device of the present invention.

Among the figure, 1 is the input unit of handwritten input character in the input device of tablet type; 2 is the stroke feature extraction unit that extracts the feature of the stroke that constitutes handwritten character from the handwritten character that input unit 1 inputs; 3 is to a plurality of Recognition object character is stored in advance according to correct stroke order writing each recognition object character i.e. constitutes the stroke characteristic dictionary of the linear stroke of standard figure and the stroke feature of non-linear stroke; 4 is the feature that extracts according to stroke feature extraction unit 2, the Each stroke is classified into the stroke classification unit of one of straight stroke, non-straight stroke and successor character stroke; 5 is the straight stroke that carries out the characteristic of the straight stroke that is classified by above-mentioned stroke classification unit 4 and constitutes the standard figure stored in the stroke feature dictionary 3 The straight line stroke corresponding connection unit of the corresponding connection of the feature; 6, according to the feature that is extracted by stroke feature extraction unit 2, the stroke that determines the existence area of each stroke of handwritten character can correspond to the connection area determination unit; 7 is to carry out by stroke classification unit 4 The non-linear stroke corresponding connection unit of the feature of the non-linear stroke of classification and the feature of the non-linear stroke that constitutes the standard figure stored in the stroke feature dictionary 3; Stroke or non-straight line stroke, and the feature of the straight line stroke or non-straight line stroke of this division is with the straight line stroke of the standard figure that constitutes the standard figure stored in the stroke feature dictionary 3 or the corresponding connection unit of the successor word stroke of the non-straight line stroke feature of the corresponding connection; 9 It is a partial stroke making unit for making partial strokes according to the strokes of subsequent characters.

10 is the character recognition unit that calculates the evaluation value of characters and recognizes handwritten characters according to the corresponding connection results obtained by the corresponding connection unit 5 of the straight line strokes, the corresponding connection unit 7 of the non-linear strokes, and the corresponding connection unit 8 of the subsequent character strokes, and 11 is the character recognition unit for controlling each unit control unit.

FIG. 2 is a flowchart showing the processing flow of the character recognition unit shown in FIG. 1 .

An example of the processing procedure will be described using a flowchart showing the processing flow of the character recognition unit in FIG. 2 .

First, in step S1, characters are written by the input unit 1 into the tablet-type input device with a pen.

FIG. 3 is a diagram showing the input pattern "亚" of handwritten characters input to the input unit 1, and 60-65 are strokes input in chronological order. In this example, writing with irregular stroke order and number of strokes is performed, and consists of six strokes.

Proceeding to step S2, the stroke feature extraction unit 2 extracts the features of each stroke constituting the input pattern of handwritten characters. In this example, stroke features of strokes 60-65 of the input pattern in FIG. 3 are extracted.

FIG. 4 is the stroke features of the input graphic extracted by the stroke feature extraction unit 2, showing the stroke features obtained from the strokes 60-65 of the input graphic in FIG. In this stroke feature, in order to quantify the strokes shown in FIG. 5 in 16 directions, the stroke direction and the direction of the unwritten imaginary stroke when actually connecting one stroke to the next stroke are calculated using the 16-direction distribution map with direction arrangement .

Then, proceeding to step S3, the stroke classifying unit 4 classifies the strokes of the input graphics according to their shapes.

FIG. 6 is a flowchart showing the processing flow of the stroke classifying unit 4 . The operation of the stroke classification unit 4 will be described in detail using the flowchart of FIG. 6 .

First, in step S10 of FIG. 6 , the stroke classifying unit 4 sequentially judges whether the current stroke is a straight stroke starting from the first stroke, and if it is a straight stroke, proceeds to step S11, and otherwise proceeds to step S13. In the input pattern of FIG. 3, five strokes 60, 61, 62, 63, and 65 are judged to be straight strokes according to the shape of the stroke feature in FIG. 4, and the process proceeds to step S11.

In step S11, the stroke classifying unit 4 extracts the directional component distribution of strokes. The direction distribution of the stroke is calculated from the direction distribution of each straight line connecting points where input strokes are sampled at appropriate intervals. Here, when the stroke classification unit 4 calculates the directional distribution, the 8-directional component distribution shown in FIG. 7 is used.

Proceed to step S12, the stroke classifying unit 4 checks whether the desired direction component is only in a certain direction, and when only in a certain direction, enter step S15, determine that it is a straight line stroke, otherwise just proceed to step S13. In the example of the input figure in FIG. 3, since all the straight strokes have only components in a certain direction, the process proceeds to step S15, where it is judged to be a straight stroke.

When the direction component is not only in a certain direction, in step S13, the stroke classifying unit checks whether the straight line strokes with multiple direction components and the strokes outside the straight line are stroke shapes that exist in the stroke feature dictionary 3. In the case of existing strokes, proceed to step S16, where it is judged as the stroke of a subsequent character. If it is a stroke that exists in the dictionary, it proceeds to step S14 and determines that it is a non-linear stroke. In the example of the input figure in FIG. 3, since the unknown stroke 64 does not exist in the dictionary, it proceeds to step S16, and determines that it is a subsequent character stroke.

Then, enter step S4 of Fig. 2, the control unit 11 sends the strokes determined as straight strokes in the stroke classification unit 4 to the straight line corresponding connection unit 5, and performs the straight line strokes of the input graphics and the straight lines of the standard figures of the stroke feature dictionary 3 Corresponding connections of graph strokes. In the example of the input pattern shown in FIG. 3 , straight line strokes 60 , 61 , 62 , 63 , and 65 are sent to the straight line stroke-corresponding connection unit 5 .

FIG. 8 is a flowchart showing the processing flow of the linear stroke correspondence connection unit. The operation of the linear stroke correspondence connection unit 5 will be described using the flowchart of FIG. 8 .

First, in step S21, the straight line stroke correspondence connection unit 5 classifies each straight line stroke into three types (long, medium, short) according to the length. The length of the stroke is obtained from the Euclidean distance between the start point and the end point of the stroke. In this example, define long strokes as greater than 70, short strokes as less than 30, and middle strokes as their intermediate values. According to the length information of the input stroke shown in the stroke feature of the input pattern in FIG. 4, strokes 60, 61, 62, and 65 are classified as long strokes, and stroke 63 is classified as short strokes.

Going to step S22, the straight line stroke corresponding connection unit 5 determines the existence area of the stroke by the stroke corresponding connection area determination unit 6 according to the stroke direction, the coordinates of the stroke and the length of the stroke.

FIG. 9 shows an example of region division used in the correspondence connection of straight line strokes. Here, an example of the case where the stroke-existing area is determined using the area that divides the character into four in the horizontal direction and the vertical direction by the circumscribing rectangle is shown.

FIG. 10 shows information for specifying an area used in the stroke-corresponding-connection specifying unit 6 . The direction of the stroke in FIG. 10 represents the value obtained by quantizing the direction from the start point to the end point of the stroke using the 16-direction map in FIG. 5 . For example, for the horizontal straight strokes in FIG. 10 , the stroke direction indicates the possibility of corresponding connection with the horizontal strokes of the standard figures in the 4, 5, 6, 12, 13, and 14 directions. In addition, the horizontal straight strokes existing in area 1 indicate the possibility of corresponding connection with the horizontal straight strokes existing in area 1 and area 2 .

As shown in FIG. 10 , the stroke-corresponding connectable area determination unit 6 judges that the distance in the direction perpendicular to the stroke direction of the input figure changes little and is relatively stable, and there are all existing areas up to the adjacent area. However, in the case of straddling a plurality of areas, all areas adjacent to the plurality of areas become existing areas.

FIG. 11 shows the existence area and the corresponding connection area of the strokes of the input pattern in FIG. 3 .

Proceeding to step S23, the straight line stroke-corresponding connection unit 5 determines the strokes of the standard figures in the stroke feature dictionary 3 to be connected to the strokes of the input figures.

FIG. 12 is a diagram showing the strokes of the standard pattern of the character "亚" stored in the stroke feature dictionary 3. As shown in FIG. FIG. 13 shows the stroke features of the standard figure in FIG. 12 stored in the stroke feature dictionary 3 .

As the condition of linkable connection, there are three conditions: 1) the shape condition of the stroke, 2) the length condition of the stroke, and 3) the condition of the linkable region of the stroke to determine the stroke of the standard figure to be connected.

Fig. 14 is a diagram showing whether the corresponding connection can be carried out according to the stroke shape, which is to determine whether the corresponding connection can be carried out with the stroke shape used in the straight line stroke corresponding connection unit 5, the non-linear stroke corresponding connection unit 7, and the subsequent word stroke corresponding connection unit 8 picture.

Specifically, the shape condition of the stroke is the corresponding relationship shown in Figure 14, that is, the stroke shape of the input figure and the stroke shape of the standard figure, the corresponding connection of the connection "○" is regarded as a corresponding connection, and the connection of "×" is regarded as a non-corresponding connect.

Also, the length condition of the stroke is that the long stroke can be connected correspondingly with the long stroke or the middle stroke, the middle stroke can be connected correspondingly with the long stroke and the short stroke, and the short stroke can be connected correspondingly with the middle stroke or the short stroke. Also, the stroke may correspond to the condition of the connection area. The stroke may correspond to the result of the connection area determining unit 6 .

In this example, the stroke features shown in FIG. 13 are stored in the stroke feature dictionary 3 for the standard figure shown in FIG. 12 . According to the stroke feature, for each stroke of the input figure, the stroke of the standard figure corresponding to the connection is determined.

In this embodiment, for each stroke of the input figure in FIG. 3 , strokes of standard figures that can be connected are determined.

FIG. 15 is a diagram showing strokes of standard figures that can be connected according to the stroke shape condition, the stroke's connectable region condition, and the stroke length condition. Here, the summarization according to each condition is performed sequentially only for the contents summarized so far.

According to Fig. 15, stroke 60 is summarized into strokes 70 and 73, stroke 61 is summed up among strokes 74, 75, stroke 62 is summed up among strokes 75, stroke 63 is summed up among strokes 71, and stroke 65 is summed up into strokes 73, 76 middle. Here, there is no other correspondence to the summed up strokes, and the summed up standard strokes cannot be used in corresponding connections of other input strokes. Here, since the strokes 71 and 75 are combined into one, the strokes 71 and 75 are excluded from the candidates of other input strokes. Therefore, stroke 61 is grouped into stroke 74, and finally, there are only two candidate strokes, stroke 60 and stroke 65.

Next, enter step S24, the straight line stroke corresponding connection unit 5 performs corresponding connection with the strokes of the input figure and the candidate strokes of the standard figure in detail, and the successful result of the corresponding connection is used as the corresponding connection result of all the straight line strokes.

Here, in the check of the detailed corresponding connection, compare the stroke direction, stroke width, stroke height, and the coordinate value of the stroke point with the predetermined corresponding connection threshold, and check whether the corresponding connection is possible and where . At this time, when the input strokes are correspondingly connected with the continuous strokes of the standard figure, the corresponding connection results of the virtual strokes are also used to check the corresponding connection.

In addition, in order to absorb the change of the strokes of the input graphics, even when the stroke order of the input graphics is different from that of the standard graphics, all the candidate strokes of the standard graphics that can be connected with the strokes of the input graphics are correspondingly connected.

Specifically, compare the stroke features of the input graphic in FIG. 4 with the stroke features of the standard graphic in FIG. 13 .

In this example, for the strokes 61, 62, and 63 of the input pattern, the candidates that have already been connected are collected into one, and since any stroke can be correctly connected, the corresponding connection is successful. For the strokes 60 and 65, since there are two candidates, the strokes for each input figure correspond to the candidate strokes connecting the standard figures. That is, the stroke 60 of the input figure is connected to the stroke 70 and the stroke 73 of the standard figure, and the stroke 65 is connected to the stroke 73 and the stroke 76 .

First, the corresponding connection processing of the stroke 60 of the input figure will be described. Corresponding connections between the stroke 60 and the strokes 70 and 73 of the standard figure are performed. Here, when the stroke 60 and the stroke 73 are connected in correspondence, the difference between the Y coordinate values of the dots always becomes large. Also, if the virtual strokes representing the positional relationship between the strokes are used to check the corresponding connection results, then as can be seen from Figure 13, in the case of stroke 73, the direction of the virtual stroke pointing to the next stroke 74 is "16". In the case of stroke 60, the stroke of the input figure correspondingly connected with the standard stroke 74 in the dictionary is stroke 61, and the virtual stroke direction of stroke 60 becomes the direction from the end point of stroke 60 to the starting point of stroke 61, so it is "13" . Therefore, the direction difference of virtual strokes is 16-13=3.

On the other hand, in the corresponding connection of stroke 60 and stroke 70, the difference between the point coordinate values and the direction difference between the virtual strokes decrease (the distance of the corresponding connection result decreases), which is a correct corresponding connection.

Next, the corresponding connection between the stroke 65 and the strokes 73 and 76 of the standard figure is performed. When the stroke 65 and the stroke 73 are connected in correspondence, the difference in the coordinate values of the dots (the difference in stroke characteristics) always becomes large. On the other hand, when the stroke 65 and the stroke 76 are connected in correspondence, the difference in the characteristics of the strokes is small, and the corresponding connection is correct.

As a result, the corresponding connection of straight line strokes is that the stroke 60 of the input figure is connected with the stroke 70 of the standard figure, the stroke 61 is connected with the stroke 74, the stroke 62 is connected with the stroke 75, the stroke 63 is connected with the stroke 71, and the stroke 65 is connected with the stroke 76 corresponding connections, summed up as a kind of corresponding connection.

As shown in this example, with the above-mentioned corresponding connection processing, even if the stroke order of the strokes of the input figure is different from that of the standard figure, the corresponding connection can be made.

Proceed among the step S25, the straight line stroke corresponding connecting unit 5 enters into step S27 under the situation that the straight line stroke of the input figure is corresponding to any stroke of the standard figure, and the corresponding connection is (OK), if there is an uncorresponding stroke ( Even if there is only one), go to step S28, and the corresponding connection with the current standard figure is "impossible" (NG).

In this example, the straight line strokes of all input graphics are correctly connected with the strokes of standard graphics.

Next, in step S5 of FIG. 2 , the control unit 11 checks the result of the corresponding connection of the straight line strokes, and if the corresponding connection is "impossible", it proceeds to step S8 to perform the corresponding connection with the next standard figure. In addition, when the corresponding connection is successful, the process proceeds to step S6, and the corresponding connection with the non-linear stroke is performed.

FIG. 16 is a flowchart showing the processing flow of the non-linear stroke correspondence connection unit 7 . Using the flowchart shown in FIG. 16 , the operation of the non-linear stroke correspondence connection unit 7 will be described in detail.

In step S30, the non-linear stroke corresponding connection unit 7 is the same as the straight stroke. For non-linear strokes, the stroke corresponding connection area determination unit 6 determines the strokes of the standard graphics that are correspondingly connected with the strokes of the input graphics. This correspondence is the same as that of straight line strokes, and the information map for specifying the area shown in FIG. 10 is used.

Here, the corresponding connection between the non-linear strokes of the input graphics and the strokes of the standard graphics is limited to the standard graphics in which the linear strokes correspondingly connected by the above-mentioned straight stroke corresponding connection unit 5 exist.

Proceeding to step S31, the non-linear stroke correspondence connection unit 7 performs correspondence connection with the stroke shape of the standard figure in addition to the stroke shape of the input figure.

First, for the strokes of the standard pattern that can be associated with each stroke of the input pattern obtained in the process of step S30, the corresponding connection is checked according to the shape of the stroke. The stroke shape correspondence connection rule uses a graph indicating whether a correspondence connection corresponding to the stroke shape in FIG. 14 is possible.

Next, the detailed stroke correspondence check is carried out similarly to the linear strokes for the strokes judged to be correspondingly connected.

Proceed to step S32, when the non-linear stroke corresponding connection unit 7 judges that the stroke of the input figure is connected with one of the strokes of the standard figure, it proceeds to step S34, and determines that the corresponding connection is successful. Here, when there is no corresponding connection, it is considered that there is no consistent stroke with the stroke of the standard figure, and it proceeds to step S33, and is judged as a follow-up character stroke.

Proceeding to step S35, the non-linear stroke corresponding connection unit 7 studies whether all non-linear strokes have been connected, and if not, proceeds to step S30 to perform the corresponding connection of the next non-linear stroke. In addition, when all the corresponding connections are completed, the corresponding connection of the non-linear stroke corresponding connection unit 7 is completed.

Next, proceed to step S7 of FIG. 2 , the control unit 11 performs the corresponding connection of the subsequent character strokes by the subsequent character stroke corresponding connection unit 8 .

FIG. 17 is a flow chart showing the processing flow of the successor character stroke correspondence connection unit 8 . Next, the operation of the successor-character-stroke-corresponding linking unit 8 will be described using the flowchart of FIG. 17 .

First, in step S50, the subsequent character stroke corresponding connection unit 8 determines the corresponding connection area of the subsequent character stroke of the input graphic by the stroke corresponding connection area determination unit 6, and obtains the standard graphic that can be correspondingly connected with the subsequent character stroke of the input graphic stroke candidates for .

Here, the corresponding connection of subsequent character strokes of the input graphics is limited to the standard graphics with straight and non-linear strokes correspondingly connected by the above-mentioned linear stroke corresponding connection unit 5 and the above-mentioned non-linear stroke corresponding connection unit 7 .

FIG. 18 is a diagram showing an example of region division used by the stroke-corresponding-connection region determining unit 6 when connecting other than straight strokes. When the stroke corresponding connectable area determination unit 6 is outside the strokes shown in the information map for corresponding connectable area determination in FIG. The stroke-corresponding connection region is targeted at an area up to one outside of the area contained in the circumscribed rectangle of the stroke of the input figure. In the example of the input pattern shown in FIG. 3, the stroke 64 is judged to be a subsequent character stroke and not the shape of the stroke shown in FIG.

FIG. 19 is an example of dividing the input figure of FIG. 3 by the region of FIG. 18 and obtaining the circumscribed rectangle 80 of the stroke 64, which represents where each stroke of the input figure exists in the region where the stroke can correspond to the connection area determination unit 6. picture. It can be seen from the figure that the circumscribed rectangle 80 of the stroke 64 is included in the areas A-H, and the corresponding connectable areas of the strokes are the areas A-L.

Next, the successor character strokes correspond to the connection unit 8 to obtain the determined strokes that can correspond to the standard graphics existing in the connection area. In the case of the stroke example of the standard figure shown in Figure 12, there are four strokes 70, 71, 72, and 73 included in the regions A-L, but the connecting unit 5 corresponding to the straight line stroke corresponds to the non-linear stroke The strokes of the standard figure connected by the connecting unit 7 to the input strokes in one-to-one correspondence are removed because the corresponding connection has been determined. Therefore, in this example, the strokes 70 and 71 are eliminated because they are determined by straight line strokes corresponding to connection units. As a result, strokes 72, 73 are selected as candidate strokes.

Proceed to step S57, whether the subsequent character stroke corresponding connection unit 8 checks whether the corresponding connection area is determined for all subsequent character strokes, if it is determined, it will proceed to step S51, if it is not determined, it will return to step S50.

In the input figure of Fig. 3, because the successor character stroke has only one, so proceed to step S51, determine the candidate stroke of the standard figure that is formed by the start stroke and the end stroke of the successor character stroke of input figure.

Concretely, at first, for the continuation point of the successor character stroke of input figure, obtain the region that its continuation point exists with the region segmentation of Fig. 18, with the region until the outside of existing region as corresponding connection region.

Next, the strokes of the standard figure corresponding to the corresponding connectable regions that exist at the start point of the subsequent character strokes of the input figure select the stroke including the start point as a candidate for the start stroke. Likewise, the strokes of the standard figure corresponding to the corresponding connectable regions that exist at the end points of the input successor strokes select the strokes including the end points as candidates for the end strokes.

In the case shown in FIG. 19 , since the starting point of the successor stroke 64 of the input graphic exists in area A, the corresponding connectable areas of the starting point are areas A, B, E, and F. Since the start points of the strokes 72 and 73, which are the candidates for the standard figure obtained in the process of step S50, are included in this area, the candidates for the start strokes are two strokes 72 and 73. Similarly, since the end point of the subsequent stroke 64 exists in the area H, the corresponding connection areas are areas C, D, G, H, K, and L. In this way, the candidates for the end point strokes are also the strokes 72 and 73 .

Proceed to step S52, whether the total number of strokes (wherein, the same stroke does not repeat counting) of the total point candidate that step S51 seeks according to the connection unit 8 checks step S51 is less than 3, in the case of less than 3, proceed to step S54, If it is greater than 3, go to step S53. In the case of the subsequent character strokes in FIG. 3, since the number of candidate strokes is 2, the process proceeds to step S54.

In step S54, the corresponding connection between the strokes of the successor characters of the input graphics and the candidate strokes of the standard graphics is performed. FIG. 20 is a flowchart showing the corresponding connection processing in step S54. Using the flow chart of FIG. 20, the corresponding connection of subsequent character strokes will be described in detail.

First, in step S60, the subsequent character stroke correspondence connection unit 8 determines the stroke order of the candidate strokes of the standard figure. Here, in step S60, the number of candidate strokes of the standard figure must be reduced to three or less. Since the succeeding characters consist of at least two strokes, there are 3C2×P2+P3=12 stroke order combinations when there are 3 stroke candidates in the standard figure.

Similarly, when there are two candidate strokes for the standard figure, there are two types.

Since there are 12 types in the worst case, it is a combination that can be processed in actual time, but the number of combinations is limited by the following two constraints.

1) subject to the candidate of the start stroke determined by step S51, the candidate of the end stroke is restricted

2) Restricted by strokes that cannot be deleted

Here, the term "undeletable" refers to only the strokes of the standard graphics corresponding to the stroke candidates of the currently input graphics. In the process of step S50, candidates for standard figure strokes that can be associated with the strokes of the input figures are identified, so the deletion stroke can be specified using this information.

In the case of the stroke 64 of the input figure, there are two candidate stroke numbers, each of which is a candidate for a start stroke and an end stroke, and since there is no deletion stroke, there are two kinds of strokes. In this example, first, the stroke order of the strokes 72 and 73 is associated with each other.

Proceeding to step S61, the subsequent character stroke corresponding connection unit 8 performs the corresponding connection between the subsequent character strokes of the input graphic and the strokes of the standard graphic. The corresponding connection is carried out with feature point values. The stroke feature point information of standard graphics only has feature points at the start point and end point when it is a straight line stroke. Therefore, the feature points stored in the stroke feature dictionary 3 are also used. In addition, the stroke feature point information of the input figure becomes the sample points expressed at appropriate intervals after preprocessing.

Known from the stroke features of the standard figure in Figure 13, the feature point information of stroke 72 has three points: the start point, the feature point in the stroke, and the end point, and the feature point information of stroke 73 has two points: the start point and the end point. Therefore, there are five stroke feature points of standard graphics. "○" in FIG. 21 indicates the stroke feature points of the obtained standard figures. In addition, "○" in FIG. 22 indicates a feature point of the subsequent character stroke 64 of the input pattern.

Next, the successor character stroke corresponding connection unit 8 performs a corresponding connection with the minimum distance between the stroke feature points of the input figure and the stroke feature points of the standard figure. Generally, the corresponding connection between the stroke A of the input graphics and the stroke B of the standard graphics is by giving the set {0, 1, ... I} of the element strings of the feature points of the stroke A of the input graphics to the stroke B of the standard graphics The mapping ω of the set {0, 1, .

Here, ω is a mapping that makes the two endpoints of stroke A and stroke B consistent, satisfying ω(0)=0, ω(1)=J (that is, the starting point and the end point are consistent), and the order of the constituent elements does not allow reversal The monotonic mapping of , so it is called stretching mapping. Also, assume that the i-th element of stroke A is ai, the j-th element of stroke B is bj, and the distance d(ai, bj) between ai and bj is the Euclidean distance between feature points ai and bj, stroke The cumulative distance from the i-th element of A to the j-th element of stroke B is g(i, j), then the recursive formula is

g(i, j)=d(ai, bi)+min{g(i-1, k)|0≤k≤j} (Formula 2)

Where g(0, 0)=d(a0, b0), g(0, j)=∞ (when j>0) holds true.

In this way, the dynamic programming method can be used to obtain the corresponding connection distance g(I, J) between the stroke A of the input figure and the stroke B of the standard figure. Here, if the k value of min{g(i-1), k|0≤k≤j} is recorded, the best corresponding connection information between feature points can be obtained. That is to say, it can be known which feature point of the stroke of the standard figure corresponds to which feature point of the stroke of the subsequent character of the input figure.

If carry out the corresponding connection of the feature point of the stroke of the standard figure of Fig. 21 and the stroke feature point (sampling point) of the input figure of Fig. 22 with dynamic programming method, then obtain the correspondence shown in the arrow shown in the evaluation value of corresponding connection and Fig. 23 Connection result.

Proceeding to step S62, the subsequent character stroke corresponding connection unit 8 uses the stroke feature extraction unit 2 to obtain the stroke features of the segmented strokes according to the obtained corresponding connection results. In order to obtain the stroke features, first use the feature points corresponding to the connection points of the strokes of the standard graphics to segment the subsequent character strokes of the input graphics, and extract the strokes enclosed by the consistent points as the segmented strokes.

Next, stroke features of the segmented strokes are extracted by the stroke feature extraction unit 2 . FIG. 24 is an example of obtaining the divided strokes 130 and 131 from the stroke 64 of the input figure.

Proceeding to step S63, the subsequent character stroke corresponding connection unit 8 checks the corresponding connections with the subsequent character strokes of the input graphic for all the stroke orders of the current candidate strokes. If the result of checking is that all stroke combinations are not used for corresponding connection, go to step S60 and perform corresponding connection for other stroke orders.

When the stroke 64 of the figure is input, the process proceeds to step S60, and the corresponding connection is made according to the stroke order of the strokes 73 and 72 . The corresponding connection processing is the same as that of the strokes 72 and 73, so the description is omitted.

This is the case of the stroke order of strokes 72 and 73, but since the distance between the stroke feature points of the input figure and the stroke feature points of the standard figure is small, it can be used as the corresponding connection result of correct partial strokes.

If all the stroke order connections are all over, just go to step S56 of Fig. 17, the subsequent word stroke corresponding connection unit 8 checks whether all subsequent word strokes have been connected correspondingly to the subsequent word strokes of the current input graphics, after finishing the corresponding connection In this case, the processing of the succeeding character stroke-corresponding connection unit 8 ends, and the process proceeds to step S8 in FIG. 2 . When the corresponding connection is not completed, the process returns to step S51 in FIG. 17 . In the input pattern of Fig. 3, because there is only one successor character stroke, so, condition is satisfied, advances to step S8.

In step S8, the control unit 11 checks whether all standard patterns in the stroke feature dictionary 3 are matched, and proceeds to step S9 when finished, and proceeds to step S4 when not finished, and performs corresponding connection with the next standard pattern.

With regard to the input pattern in FIG. 3 , since the correspondence connection with all the standard patterns has been completed, the process proceeds to step S9.

In step S9, the control unit 11 sends the corresponding connection results obtained by the linear stroke corresponding connection unit 5, the non-linear stroke corresponding connection unit 7, and the subsequent character stroke corresponding connection unit 8 to the character evaluation unit 10.

The character evaluation unit 10 obtains the evaluation value of the characters according to the sent corresponding connection results, and selects the recognition target character (standard pattern) with the smallest distance as the final recognition result.

The evaluation value D of the character is obtained by the following formula 3.

D＝(Wd×Dd+Ww×Dw+Wh×Dh+Wvw×Dvw+Wvh×Dvh+Wvd×Dvd)/k (Formula 3)

Here, the normalized distance of the stroke direction is represented by Dd, the normalized distance of the stroke width is represented by Dw, the normalized distance of the stroke height direction is represented by Dh, and the normalized distance of the virtual stroke width is represented by Dvw. The height normalized distance is represented by Dvh, the normalized distance of the imaginary stroke direction is represented by Dvd, Wd represents the number of stroke direction, Ww represents the number of stroke width, Wh represents the number of stroke height, Wvw represents the number of imaginary stroke width , Wvh represents the number of imaginary stroke height, Wvb represents the number of imaginary stroke direction, and k represents the number of strokes of the current standard graphics. Also, Wd+Ww+Wh+Wvw+Wvh+Wvd=1.

But, under the situation that there are multiple corresponding connection results of straight line strokes, non-straight line strokes and successor word strokes, consider not producing the stroke of the non-corresponding standard figure that does not correspond to any stroke of the input figure and the stroke of a standard figure. Inconsistencies between stroke correspondences with strokes of multiple input shapes.

Next, the partial stroke extraction unit 9 will be described. Fig. 25 is an example of an input pattern including the character "言" written with four strokes and seven strokes.

Next, the operation of the partial stroke extraction unit 9 will be described by taking the input pattern shown in FIG. 25 as an example. In this example, the strokes 141, 142, and 143 are connected correspondingly by using the straight line stroke corresponding connection unit 5 and the non-linear stroke corresponding connection unit 7, and the description thereof is omitted.

Stroke 140 is a successor character stroke, so it enters step S7 of FIG. 2, and the successor character stroke corresponding connection unit 8 performs corresponding connection.

At first, advance to step S50 of Fig. 17, subsequent character stroke corresponding connection unit 8 determines the corresponding connection region of the subsequent character stroke of input figure by stroke corresponding connection area determination unit 6, seeks the standard that can be correspondingly connected with subsequent character stroke Stroke candidates for graphics.

The stroke-corresponding connectable area determining unit 6 determines a correspondingly connectable area using the area division information of FIG. 18 . It is possible to target the area outside the area of the circumscribed rectangle including the strokes of the input figure corresponding to the link area. FIG. 26 is an example of performing region division as shown in FIG. 18 on the input pattern shown in FIG. 25 . It can be seen from FIG. 26 that the circumscribed rectangle 144 of the stroke 140 includes areas A-L. As a result, the stroke-corresponding connectable regions of the stroke 140 are all the regions AP.

The corresponding connection unit 8 of the successor character stroke obtains the strokes of the standard graphics existing in the determined corresponding connection area. Fig. 27 is the standard figure of character " word ", be that all strokes are contained in the area A～P, but because wherein utilize straight line stroke corresponding connection unit 5 and non-linear stroke corresponding connection unit 7 and the stroke of input figure one by one The corresponding connection of the standard graphic stroke corresponding to the connection has been determined, so it is removed. In this example, since the strokes 154 and 155 are determined by the straight-line stroke-corresponding connection unit 5, and the stroke 155 is determined by the non-straight-line stroke-corresponding connection unit 7, they are removed. As a result, four strokes 150, 151, 152, and 153 are selected as candidate strokes.

Then, enter the step S57 of Fig. 17, whether successor character stroke corresponding connection unit 8 checks to all successor character strokes and determine the corresponding connection area; If determined, just proceed to step S51, if undetermined, just proceed to step S50. In the input figure of Fig. 25, because there is only one succeeding character stroke, so, advance to step S51, determine the candidate stroke of the standard figure that is formed by the start stroke of succeeding character stroke, the end stroke.

From Fig. 26, since the starting point of the stroke 140 of the input figure exists in the region B, the corresponding connection regions become regions A, B, C, D, F, G, and the starting point includes the stroke 150 of the standard figure in this region , 151, 152 become candidates for starting strokes. Likewise, strokes 152, 153 become candidates for ending strokes.

Proceed to step S52, and the stroke candidate number of subsequent word stroke corresponding connection unit 8 checks standard figure, but in this example, because the candidate character stroke number is 4, so, proceed to step S53.

In step S53, the subsequent character stroke corresponding connection unit 8 uses the partial stroke extraction unit 9 to extract partial strokes from the subsequent character strokes of the input graphic.

The operation of the partial stroke extraction unit 9 will be described using the flowchart shown in FIG. 28 .

In step S70, the partial stroke extracting unit 9 extracts the inflection point of the stroke of the input figure as a segment candidate point of the partial stroke, and uses the starting point of the partial stroke as the starting point of the stroke of the input figure.

Fig. 29 is a diagram showing candidates for breaking points of subsequent character strokes of an input pattern. In FIG. 29 , 160 to 165 are partial stroke division candidate points of the stroke 140 of the input figure, 160 is the starting point of the partial stroke, and 167 is the end point of the partial stroke.

Proceed to step S71, the partial stroke extracting unit 9 reversely searches for a breakpoint from the stroke end point of the input graphic, and obtains a breakpoint whose number of candidate strokes is less than 3 from the stroke starting point of the current part to the stroke of the breakpoint, and regards the breakpoint as a part The end point of the stroke. In the example of Fig. 29, the partial strokes from the starting point 166 to the breaking point candidate points 160, 161, 162, 163, 164, 165 are selected at first, and the partial strokes exist in the regions A, B, C, D, E, F , G, I, J, and K, so the correspondingly connected areas are all areas, and the number of candidate strokes 4 is still 4, so the previous break candidate point 164 of the break candidate point 165 is entered. Similarly, move the sequential break point along the direction of the starting point, obtain some strokes, and extract its candidate strokes. In this example, since the number of candidate strokes for a partial stroke from the start point 166 to the segment candidate points 160, 161, and 162 is three, this stroke is selected as a partial stroke.

Proceeding to step S72, the successor character stroke corresponding connection unit 8 performs corresponding connection between the obtained partial strokes and candidate strokes.

FIG. 31 is a flowchart showing the flow of processing in step S72. The processing of step S72 will be described using the flowchart of FIG. 31 .

In step S80 , the circumscribing rectangles of the partial strokes and the circumscribing rectangles of the candidate strokes are used to count candidates for corresponding connection. Specifically, a candidate stroke including a circumscribed rectangle is selected in an area of a circumscribed rectangle of a partial stroke developed by a predetermined threshold. 180 in FIG. 30 is the circumscribed rectangle of the expanded partial stroke. In this example, the stroke 152 among the candidate strokes 150, 151, 152 is removed from the candidates.

Proceeding to step S81, the successor character stroke corresponding connection unit 8 determines the stroke order of the candidate strokes of the standard figure in the same way as the case where the number of candidate strokes is less than 3.

Here, when the current partial stroke includes the starting point of the stroke of the succeeding character of the input pattern or includes the candidate and end point of the starting stroke, the determination of the stroke order combination also uses the information of the candidate of the ending stroke. There are two candidate strokes for the current partial strokes, and any stroke satisfies the condition of the start stroke, so there are two stroke orders.

Here, first, the corresponding connection of the stroke order of the strokes 150 and 151 is performed.

Proceeding to step S82, the subsequent character stroke corresponding connection unit 8 performs corresponding connection between partial strokes and candidate strokes of standard graphics. The method of corresponding connection is the same as step S61 of FIG. 20 (same as the above-mentioned successor character stroke corresponding connection). In step S61 of FIG. 20, the strokes 150 and 151 are straight line strokes, and their respective start points and end points become feature points, and the feature points of part of the strokes are obtained by using dynamic norms.

Proceed to step S83, the successor character stroke corresponding connection unit 8 divides some strokes according to the obtained corresponding connection result, and the stroke feature extraction unit 2 obtains the stroke features of the divided strokes. FIG. 32 is an example of obtaining segmented strokes 190 and 191 from the partial strokes in FIG. 30 .

Proceeding to step S84, the successor character stroke corresponding connection unit 8 checks whether all stroke orders of the current candidate strokes have been connected with some strokes of the input graphic. As a result of the check, if it is not performed for all combinations, proceed to step S81, and perform corresponding connections for other stroke orders.

In this example, proceed to step S81 to perform corresponding connection of strokes 151 and 150 in other stroke orders. The corresponding connection processing is the same as that of the strokes 150 and 151, so the description is omitted. Here, the stroke order of the strokes 150 and 151 is adopted as the corresponding connection result of the correct partial strokes because the distance between the stroke feature points of the input figure and the stroke feature points of the standard figure decreases.

Then, proceed to step S73 of Fig. 28, the subsequent word stroke corresponding connection unit 8 judges whether the corresponding connection of the current subsequent character stroke ends, advances to step S75 when ending, advances to step S74 when not ending.

In the example of FIG. 29, since it is not finished, it progresses to step S74. In step S74, the end point of the current partial stroke is used as the starting point of the next partial stroke. Here, the segment candidate point 162 in FIG. 29 becomes the starting point of the next partial stroke.

Proceed to step S71, the successor character stroke corresponding connection unit 8 uses the partial stroke extraction unit 9 to make the next partial stroke. Specifically, for a new partial stroke, the partial stroke extracting unit 9 reversely obtains a break point from the end point of the partial stroke, and obtains a break point whose number of candidate strokes from the starting point of the current partial stroke to the break point is less than 3 , and use the breaking point as the end point of the partial stroke.

In the example of FIG. 29, since the specified strokes 150 and 151 are excluded from the candidates, any one of the strokes 152 and 153 is selected as a candidate. Therefore, all remaining partial strokes composed of the segment candidate points 162, 163, 164, and 165 and the end point 167 of the partial stroke are selected.

Proceeding to step S72, the successor character stroke corresponding connection unit 8 performs corresponding connection between the obtained partial strokes and candidate strokes.

The operation of partial stroke correspondence connection will be described using the flowchart in FIG. 31 .

In step S80, candidates for corresponding connections are inserted using the circumscribing rectangles of some strokes and the circumscribing rectangles of each candidate stroke. In this example, the candidate strokes 152, 153 become the objects of the corresponding connection that both satisfy the conditions.

Next, proceed to step S81, and the subsequent character strokes correspond to the connection unit 8 to determine the stroke order. Since there are two candidate strokes for the current partial stroke, and any stroke satisfies the condition of the end stroke, there are two kinds of strokes.

Here, firstly, the corresponding connection of the order of strokes 152 and 153 is performed.

Going to step S82, the subsequent character stroke corresponding connection unit 8 performs corresponding connection between partial strokes and candidate strokes of standard graphics. The corresponding connection method is the same as step S61 of FIG. 20. Since the strokes 152 and 153 are all straight strokes, each point is a feature point, and the corresponding connection with the feature points of some strokes is carried out with the dynamic specification method.

Here, since the start point 162 of the current partial stroke coincides with the end point of the partial stroke up to this point, there may be a virtual stroke component that is not actually written between the strokes. Therefore, the corresponding connection starting from the division candidate point 162 and the corresponding connection starting from the next division candidate point 163 are performed, and the one with the smaller distance is used as the final result. In this example, the corresponding connection starting from the division candidate point 163 is selected.

Proceed to step S83, the corresponding connection unit 8 of the subsequent character strokes divides the partial strokes from the obtained corresponding connection result, and uses the stroke feature extraction unit 2 to obtain its stroke features. Fig. 33 is an example of obtaining the segmented strokes 192 and 193 from partial strokes.

Proceeding to step S84, the successor character stroke corresponding connection unit 8 checks whether all stroke orders have been connected with some strokes of the input graphics for the current candidate strokes. When the checking result is that all combinations have not been performed, proceed to step S81 to perform corresponding connections to other stroke orders.

In this example, step S81 is performed to perform corresponding connection with the stroke order of strokes 153 and 152. The corresponding connection processing is the same as that of stroke order 152 and 153, so the description is omitted. Here is the case of the stroke order 152, 153, but because the distance between the stroke feature of the input figure and the stroke feature point of the standard figure is reduced, it is adopted as the corresponding connection result of the correct partial strokes.

Then, proceed to step S73 of Fig. 28, whether the corresponding connection unit 8 of subsequent character strokes judges whether the corresponding connection of current subsequent character strokes ends, just proceeds to step 75 when ending, just proceeds to step S74 when not ending. In this example, since it has already ended, it goes to step S75.

In step S75, the subsequent character stroke corresponding connection unit 8 calculates the corresponding connection result of the subsequent character strokes according to the corresponding connection results of each partial stroke. Here, since each partial stroke corresponds to the stroke connecting two standard figures, the stroke in which the two are arranged in the order of the partial strokes is adopted as the final successor character stroke.

In the case of this example, although there are strokes without corresponding connections in the strokes of the standard figure, if there are strokes without corresponding connections, the strokes that are connected with the continuous partial strokes after including the strokes without corresponding connections The corresponding connection and the distance of the corresponding connection become smaller, and the stroke is adopted. At this time, the number of strokes of the standard graphic inserted between each partial stroke is only one.

Through the above-mentioned processing, the stroke-corresponding connection of partial strokes extracted from the subsequent character strokes is completed, and the final character evaluation value is calculated by the character evaluation unit 10 to determine the recognition result.

Also, in this embodiment, the coordinate information of feature points is used in the corresponding connection of subsequent character strokes, but it is also possible to sample the strokes at a certain interval and use the direction codes between the sampling points for corresponding connection.

Also, in this embodiment, in order to determine that the stroke can correspond to the connecting region, what is used is to divide the character into 4*1, 1*4, 4*4 regions (representing the division regions of the horizontal direction*vertical direction respectively), However, the number of divisions may be any number.

Also, in this embodiment, the segmentation candidate points of subsequent character strokes are used as inflection points, but points sampled at regular intervals may also be used.

As above, according to this embodiment, for the strokes that are not followed by writing, only the stroke features are used for corresponding connection, only for the strokes that are not correspondingly connected, the fine feature point information is used for corresponding connection, and only the characteristics of the strokes are used for final evaluation, so , can perform character recognition at high speed, and can realize a strong recognition method for character deformation at the same time.

In addition, not only there are no subsequent strokes, but also the corresponding connection of the stroke order is performed for the subsequent strokes, and character recognition can be performed with high accuracy.

Also, since the strokes to be written next are divided into partial strokes and connected accordingly, the corresponding connection can be carried out in consideration of changes in the stroke order without increasing the processing time.

In addition, handwritten characters, that is, input strokes, are classified into straight line strokes, non-straight line strokes or subsequent word strokes, and corresponding connections are made in the order of the corresponding connections of straight line strokes, the corresponding connection of non-straight line strokes, and the corresponding connection of subsequent word strokes. Since the subsequent corresponding connection processing is not performed on standard patterns that fail to be connected, efficient corresponding connection processing can be performed without performing corresponding connection with standard patterns different from the correct solution, and high-speed character recognition can be performed.

Also, since the corresponding connection of subsequent character strokes is performed according to the feature points of the start point and end point of the stroke, the corresponding connection can be performed at high speed, and the capacity of the stroke feature dictionary can be kept small.

The present invention is constituted as described above, therefore, the effects described below are achieved.

In the first or fifth direction of the present invention, according to the stroke shape of the handwritten characters (input graphics) written on the input board, the strokes are classified into linear strokes, non-linear strokes and subsequent word strokes, and the characteristics of each stroke and The stroke features of standard graphics (characters to be recognized) are connected correspondingly, and even if the input graphics vary in stroke order and number of strokes, high-speed and high-precision character recognition can be performed.

In the second or sixth aspect, since the existence area of each stroke of the input pattern is determined, and the strokes of the standard pattern corresponding to the existence area of each stroke are correspondingly connected with the strokes of the input pattern, character recognition can be performed at high speed.

In the third or seventh aspect, the non-linear strokes of the input graphics and the strokes of the standard graphics are correspondingly connected to the correspondingly connected standard graphics by direct stroke-corresponding connection, so character recognition can be performed at high speed.

In the fourth or eighth aspect, the strokes of the subsequent characters of the input graphics and the strokes of the standard graphics are connected correspondingly to the correspondingly connected standard graphics by corresponding connection of linear strokes and non-linear strokes, so that character recognition can be performed at a high speed.

Claims (8)

1. character recognition device comprises:

Input block, online input handwriting;

The stroke feature dictionary, to a plurality of identifying object literal, storage in advance constitutes the straight line stroke of each identifying object literal and the feature of non-rectilinear stroke;

The stroke feature extraction unit is from being made of the feature of the stroke of this handwriting the above-mentioned handwriting extraction of above-mentioned input block input; It is characterized in that,

The stroke taxon, according to the feature of extracting by the stroke feature extraction unit, with each stroke classification of above-mentioned handwriting be above-mentioned straight line stroke, above-mentioned non-rectilinear stroke or can not be with the consecutive word stroke of the corresponding connection of stroke that constitutes above-mentioned identifying object literal some;

The corresponding linkage unit of straight line stroke carries out the corresponding connection by the feature of the straight line stroke that constitutes the identifying object literal of storing in the feature of the straight line stroke of this stroke taxon classification and the above-mentioned stroke feature dictionary;

The corresponding linkage unit of non-rectilinear stroke carries out the corresponding connection by the feature of the non-rectilinear stroke that constitutes the identifying object literal of storing in the feature of the non-rectilinear stroke of this stroke taxon classification and the above-mentioned stroke feature dictionary;

The corresponding linkage unit of follow-up word stroke, to be divided into above-mentioned straight line stroke or above-mentioned non-rectilinear stroke by the follow-up word stroke of above-mentioned stroke taxon classification, the corresponding connection of the straight line stroke of the formation identifying object literal of storing in the feature of carrying out this straight line stroke cut apart or non-rectilinear stroke and the above-mentioned stroke feature dictionary or the feature of non-rectilinear stroke;

The literal evaluation unit connects the result according to the correspondence by the corresponding linkage unit of above-mentioned straight line stroke, the corresponding linkage unit of above-mentioned non-rectilinear stroke or the corresponding linkage unit gained of above-mentioned follow-up word stroke, discerns above-mentioned handwriting.

2. the character recognition device of claim 1 record is characterized in that: also comprise the corresponding determining unit that connects of stroke, according to the feature of being extracted by above-mentioned stroke feature extraction unit, determine the existence district of each stroke of above-mentioned handwriting;

The corresponding linkage unit of above-mentioned straight line stroke, above-mentioned non-rectilinear connect the corresponding linkage unit of stroke or the corresponding linkage unit of above-mentioned follow-up word stroke to carry out distinguishing corresponding, as to constitute the feature of the straight line stroke that is stored in the identifying object literal in the above-mentioned stroke feature dictionary or non-rectilinear stroke corresponding connection with being existed by above-mentioned each stroke of can corresponding bonding pad determining unit determining.

3. the character recognition device of claim 1 record, it is characterized in that: the corresponding linkage unit of above-mentioned non-rectilinear stroke is to carrying out correspondence connection by the corresponding above-mentioned identification literal that connects of the corresponding linkage unit of above-mentioned straight line stroke.

4. the character recognition device of claim 1 record, it is characterized in that: the corresponding linkage unit of above-mentioned subsequent words stroke is to carrying out correspondence connection by corresponding linkage unit of above-mentioned straight line stroke or the corresponding above-mentioned identifying object literal that connects of the corresponding linkage unit of above-mentioned non-rectilinear stroke.

5. the character recognition method of a character recognition device, wherein character recognition device comprises the stroke feature dictionary of a plurality of identifying object literal being stored in advance the feature of the straight line stroke that constitutes each identifying object literal and non-rectilinear stroke, comprising:

The input step of online input handwriting;

Extract the stroke feature extraction step of the feature of the stroke that constitutes this handwriting from above-mentioned handwriting by above-mentioned input step input; It is characterized in that,

According to the feature of extracting by this stroke feature extraction step, with each stroke of above-mentioned handwritten text be categorized into above-mentioned straight line stroke, above-mentioned non-rectilinear stroke and can not with one stroke classification step in the follow-up word stroke of the corresponding connection of stroke that constitutes above-mentioned identifying object literal;

Carry out by the straight line stroke feature of this stroke classification step classification with constitute above-mentioned stroke feature dictionary in the corresponding Connection Step of straight line stroke of corresponding connection of feature of straight line stroke of the identifying object literal stored;

Carry out by the feature of the non-rectilinear stroke of above-mentioned stroke classification step classification with constitute above-mentioned stroke feature dictionary in the corresponding Connection Step of non-rectilinear stroke of corresponding connection of feature of non-rectilinear stroke of the identifying object literal stored;

The feature of will become above-mentioned straight line stroke or above-mentioned non-rectilinear stroke by the follow-up word stroke segmentation of above-mentioned stroke classification step classification, carrying out straight line stroke that this is cut apart or non-rectilinear stroke with constitute above-mentioned stroke feature dictionary in the corresponding Connection Step of follow-up word stroke of the corresponding connection of feature of the straight line stroke of the identifying object literal stored or non-rectilinear stroke;

And the literal evaluation procedure of discerning above-mentioned handwriting according to the correspondence connection result of the corresponding Connection Step of above-mentioned straight line stroke, the corresponding Connection Step of above-mentioned non-rectilinear stroke or the corresponding Connection Step gained of above-mentioned follow-up word stroke.

6. the character recognition method of claim 5 record is characterized in that, comprises according to the feature of being extracted by above-mentioned stroke feature extraction step determining that the stroke in existence district of each stroke of above-mentioned handwriting can corresponding bonding pad determining step;

Above-mentioned straight line stroke Connection Step, above-mentioned non-rectilinear stroke Connection Step or the corresponding Connection Step of above-mentioned follow-up word stroke carry out with the existence district of above-mentioned each stroke of can corresponding bonding pad determining step determining corresponding, constitute be stored in the corresponding above-mentioned stroke feature dictionary in the existence district of each stroke in the straight line stroke of identifying object literal or the corresponding of feature of non-rectilinear stroke be connected.

7. the character recognition method of claim 5 record, it is characterized in that: the corresponding Connection Step of above-mentioned non-rectilinear stroke is to carrying out correspondence connection by the corresponding above-mentioned identifying object literal that connects of the corresponding step of above-mentioned straight line stroke.

8. the character recognition method of claim 5 record, it is characterized in that: the corresponding Connection Step of above-mentioned follow-up word stroke is to carrying out correspondence connection by corresponding step of above-mentioned straight line stroke or the corresponding above-mentioned identifying object literal that connects of the corresponding Connection Step of above-mentioned non-rectilinear stroke.

Images