CN102855498B - Character identifying method and device - Google Patents

Abstract

The present invention relates to a character recognition method and device, wherein the character recognition method includes: using the cross ratio to select a quadruple group composed of four points in order from the convex hull polygon of the character to be recognized; transforming the character to be recognized to the selected In the perspective-invariant coordinate system determined by the quadruple group; from the characters to be recognized after the transformation, extract features to obtain the feature vector of the characters to be recognized; in the pre-stored character category table, search and obtain the characteristics of the characters to be recognized For the records matched by the vector, vote for the character category corresponding to the found record; repeat the above steps for a predetermined number of times for different quadruples on the convex hull polygon of the character to be recognized; and determine the character category with the largest number of votes as the character Recognition results.

Description

Character recognition method and device

technical field

The present invention relates to the field of pattern recognition, more particularly to a character recognition method and device.

Background technique

Recognition of characters under large perspective deformation is a very important topic, because recognizing characters with perspective deformation is the basis of many practical applications, and characters with perspective deformation widely exist in our daily life, for example, characters in real scenes Character recognition.

To solve this problem, a basic approach is to first rectify the perspective-distorted image to an orthographic image, and then perform traditional OCR (Optical Character Recognition) recognition on the rectified image. But this method is limited by specific applications, such as requiring the existence of outlines and character lines, as well as some structures. So technicians start to focus on recognizing each single character.

Geometric Hashing (GH) is a general-purpose model-based object recognition algorithm, which is also applicable when the object has undergone various transformations and only part of the information exists. The advantages of geometric hashing algorithms are that they can be easily parallelized and the ability to work when only partial information exists. Therefore, the geometric hash algorithm is used in object recognition of affine transformation and three-dimensional object recognition.

Approximating the transmission model using an affine transformation model is a traditional method for recognizing perspective deformed characters. M.Iwamura, T.Tsuji, A.Horimatsu and K.Kise et al. improved the geometric hash algorithm in the article entitled "Realtime camera-based recognition of characters and pictograms" published by ICDAR in 2009, and proposed a A real-time algorithm for recognizing characters captured by a camera. The algorithm uses an affine model, and in order to construct an affine invariant coordinate system, three coordinate points (3-tuples) are needed as the basis of the coordinate system. Affine invariants, ie, center of gravity and area ratio, are used to reduce the degrees of freedom of 3-tuples. However, since the affine model can only be considered as an approximation of the perspective model when the size of the object is sufficiently smaller than the distance between the object and the camera, that is, the perspective distortion is small, and when the perspective distortion is large, the above approximation no longer holds true. Therefore, in order to recognize characters under large perspective deformation, new methods need to be developed.

Another existing method for identifying perspective deformed characters is to form a cross ratio spectrum for each character, and identify characters by comparing the cross ratio spectrum of the current character with the cross ratio spectrum of the template character (see Linlin Li and Chew Lim Tan published an article entitled "Character recognition under severe perspective distortion" in ICPR in 2008). The disadvantage of this method is that the time required for character recognition increases linearly with the number of categories of characters to be recognized. Therefore, the use of this method in applications with many types of characters to be recognized is limited.

Contents of the invention

According to the first aspect of the present invention, a kind of character recognition method is provided, comprising: utilize the intersection ratio to select the quaternion of ordered four points from the convex hull polygon of character to be recognized; In the perspective-invariant coordinate system determined by the selected quadruple; extract features from the converted character to be recognized to obtain the feature vector of the character to be recognized; search and obtain the character to be recognized in the pre-stored character category table For the records matched by the eigenvector of the found record, vote for the character category corresponding to the found record; repeat the above steps for a predetermined number of times for different quadruples on the convex hull polygon of the character to be recognized; and determine the character category that will obtain the largest number of votes is the character recognition result.

According to a second aspect of the present invention, there is provided a character recognition device, comprising: a selection unit configured to select a quadruple composed of ordered four points from the convex hull polygon of the character to be recognized by using the cross ratio; The transformation unit is configured to transform the character to be recognized into the perspective invariant coordinate system determined by the selected quadruple; the extraction unit is configured to extract features from the transformed character to be recognized to obtain the character to be recognized The feature vector of the character; the voting unit is configured to search the record matched with the obtained character vector of the character to be recognized in the pre-stored character category table, and vote for the character category corresponding to the found record; repeat control A unit configured to repeat the above steps a predetermined number of times for different quadruples on the convex hull polygon of the character to be recognized; and a determining unit configured to determine the character category that has received the most votes as the character recognition result.

In addition, embodiments of the present invention also provide computer programs for implementing the above methods.

In addition, embodiments of the present invention also provide at least a computer program product in the form of a computer-readable medium, on which computer program codes for implementing the above method are recorded.

Through the present invention, characters under large perspective deformation can be recognized, and compared with the traditional method using an affine model, the recognition rate of the present invention is higher.

In addition, the method of the present invention can still perform identification when some information is missing.

The method of the present invention can also distinguish and recognize characters of different fonts.

These and other advantages of the present invention will become more apparent through the following detailed description of the preferred embodiments of the present invention with reference to the accompanying drawings.

Description of drawings

The above and other objects, features and advantages of the present invention will be more easily understood with reference to the following description of the embodiments of the present invention in conjunction with the accompanying drawings. The components in the drawings are only to illustrate the principles of the invention. In the drawings, the same or similar technical features or components will be denoted by the same or similar reference numerals. In the attached picture:

Fig. 1 shows a flow chart of a method for recognizing characters according to an embodiment of the present invention;

Figure 2A shows the convex hull polygon and inner contour of the character 'H';

Figure 2B shows a diagram of the intersection of the line segment between two points on the convex hull polygon of the character 'H' and the inner contour;

Figure 2C shows a diagram of the character 'H' with perspective distortion;

Fig. 2D shows a diagram in which the line segment between two points on the convex hull polygon of the character 'H' intersects with the sawtooth on the inner contour of the character;

Fig. 3 A shows the figure that character ' H' in Fig. 2B and 2C is transformed under the perspective invariant coordinate system;

FIG. 3B shows a diagram of selected invalid quadruples transformed into a perspective-invariant coordinate system;

FIG. 4A shows a diagram of dividing the perspective invariant coordinate system into 4*4 squares;

Figure 4B shows a normalized histogram;

Figure 4C shows a diagram of a record in the hash table;

Figures 5A-5C show views of damaged and missing characters, respectively, under large perspective distortion;

FIG. 6 shows a block diagram of a device for recognizing characters according to an embodiment of the present invention;

Fig. 7 shows the composition block diagram of the extracting unit in the character recognition device; And

Fig. 8 shows a schematic block diagram of a computer that can be used to implement the method and device according to the embodiments of the present invention.

detailed description

Embodiments of the present invention will be described below with reference to the drawings. Elements and features described in one drawing or one embodiment of the present invention may be combined with elements and features shown in one or more other drawings or embodiments. It should be noted that representation and description of components and processes that are not related to the present invention and known to those of ordinary skill in the art are omitted from the drawings and descriptions for the purpose of clarity.

Fig. 1 shows a flowchart of a method for recognizing characters according to an embodiment of the present invention.

First, in step S102, an orderly four-tuple group composed of four points is selected from the convex hull polygon of the character to be recognized by using the intersection ratio.

The basic concept of Cross Radio is as follows: if P ₀ , A, B, and P ₂ are collinear, then the Cross Radio is defined as follows:

$\mathrm{<span\; class="notranslate">\; Cr</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; A</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; B</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; B</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; A</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; A</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; B</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; ,</span>$

Among them, D() represents the distance between two points.

Those skilled in the art know that the cross ratio is an invariant variable in perspective deformation, that is, under any perspective deformation, the cross ratio Cr(P ₀ , A, B, P ₂ ) remains unchanged.

Referring to Figures 2A-2D below, taking the character 'H' as an example, how to select four orderly points on the convex hull polygon of the character to be recognized will be described in detail.

In the present invention, a convex-hull polygon is defined as follows: a convex-hull polygon is a polygon whose vertices are the convex hull of a character. For example, in FIG. 2A, the convex hull polygon of the character 'H' is represented by a solid line, while the inner contour of the character 'H' is represented by a dashed line. For the two points on the convex hull polygon of the character 'H' and the two intersection points of the line segment between the two points and the inner contour of the character 'H', the intersection ratio value remains unchanged under the perspective deformation, wherein, when the intersection point is more When there are two, the first two intersection points are selected to calculate the cross ratio value. For example, in Figure 2B, P ₀ , P ₁ , P ₂ and P ₃ are the points on the convex hull polygon of the character 'H', A and B are the front of the line segment P ₀ P ₂ and the inner contour of the character 'H' Two intersection points, C and D are the first two intersection points of the line segment P ₁ P ₃ and the inner contour of the character 'H'. Point P ₀ on the convex-hull polygon of the selected character 'H' can determine P ₂ points on the convex-hull polygon, so that P ₀ , A, B, and the intersection ratio Cr _{(P 0 ,} A, B, P ₂ ) is closest to a predetermined value.

Since the inner contour of the character may appear jagged, the intersection of the line segment between two points on the convex hull polygon and the inner contour of the character may be the intersection with the sawtooth on the inner contour of the character, such as B shown in Figure 2D ₁ and _B2 . _In this case, since the distance between B1 and _B2 is very short, the calculated cross ratio value is large. In order to avoid intersecting with the sawtooth on the inner outline of the character, the point selection criterion cannot choose a too large cross ratio value. According to the inventor's observations in practice, generally speaking, the cross ratio value should be in the scope of (1, 2], and adopting a cross ratio value close to 1.5 can obtain better results. Therefore in practice, the predetermined value of the cross ratio value Usually choose 1.5.

As mentioned above, point P ₀ can be selected arbitrarily, and point P ₂ can be determined so that P ₀ , A, B, P ₂ satisfy |1.5-Cr(P ₀ , A, B, P ₂ )|minimum; then select P ₁ point, where P ₀ , P ₁ , and P ₂ are arranged counterclockwise, and P ₃ points can be determined such that P ₁ , C, D, and P ₃ satisfy |1.5-Cr(P ₁ , C, D, P ₃ ) |Minimum, this completes the selection of ordered four points from the convex hull polygon of the character to be recognized. Those skilled in the art can understand that P ₀ , P ₁ , and P ₂ can also be arranged clockwise.

Fig. 2C is a diagram showing perspective deformation of the character 'H' in Fig. 2B, where points P ₀ ′, P ₁ ′, P ₂ ′, and P ₃ ′ correspond to points P ₀ and P ₁ in Fig. 2B respectively , P ₂ and P ₃ , A', B' and C', D' are line segment P ₀ ′P ₂ ′ and line segment P ₁ ′P ₃ ′ and the first two internal contours of 'H' with perspective deformation intersection, so A', B', C', D' correspond to A, B, C, D in Fig. 2B, respectively. Since the cross ratio is a perspective invariant, the cross ratio Cr _{(P 0 ′} , _A ′, B′, P ₂ ′) and the cross ratio Cr(P ₁ ′, C′, D′, P ₃ ′) of the points P ₁ ′, C′, D′, P ₃ ′ remain unchanged. That is to say:

Cr(P ₀ ', A', B', P ₂ ') = Cr(P ₀ , A, B, P ₂ );

Cr(P ₁ ', C', D', P ₃ ') = Cr(P ₁ , C, D, P ₃ ).

According to the above method, given the point P ₀ (P ₀ ′), the point P ₂ (P ₂ ′) can be determined, no matter what kind of perspective transformation occurs; similarly, given the point P ₁ (P ₁ ′), the The P ₃ (P ₃ ′) point can be determined. The quadruples selected in this way have only two degrees of freedom, thus reducing the amount of computation.

Next, in step S104, the character to be recognized is transformed into the perspective invariant coordinate system determined by the selected quadruple.

In the present invention, the perspective invariant coordinate system is defined as a square, the side length of the square is L, and the center is (xc, yc). In the case of selecting the quaternion P ₀ , P ₁ , P ₂ , P ₃ on the character 'H' in Figure 2B, the diagram of character 'H' transformed to the perspective invariant coordinate system is shown in Figure 3A Show. The four points P ₀ , P ₁ , P ₂ , and P ₃ in Fig. 2B are respectively mapped to I ₀ (xc-l, yc+l), I ₁ (xc-l , yc-l), I ₂ (xc+l, yc-l), I ₃ (xc+l, yc+l), where, l=α*L/2, α∈[0,1], that is, I ₀ , I ₁ , I ₂ and I ₃ fall within the square. Since Figure 2C is the perspective deformation of the character 'H' in Figure 2B, in the case of selecting the corresponding quadruple P ₀ ′, P ₁ ′, P ₂ ′, P ₃ ′, the character 'H in Figure 2C The diagram transformed into the perspective-invariant coordinate system is also shown in Figure 3A, and the points P ₀ ′, P ₁ ′, P ₂ ′, and P ₃ ′ in Figure 2C are also mapped to the perspective in Figure 3A respectively I ₀ , I ₁ , I ₂ and I ₃ in the invariant coordinate system.

The four corresponding pairs of P ₀ , P ₁ , P ₂ , P ₃ (P ₀ ′, P ₁ ′, P ₂ ′, P ₃ ′) and I ₀ , I ₁ , I ₂ and I ₃ can determine a unique perspective Transformation matrix, according to the perspective transformation matrix, the character to be recognized can be transformed into a perspective invariant coordinate system.

The method of determining the perspective transformation matrix is already known, and the relevant literature includes Multiple View Geometry in Computer Vision. Richard Hartley and Andrew Zisserman, Cambridge University Press, 2004, and will not be repeated here.

Preferably, after transforming the character to be recognized into the perspective invariant coordinate system, the effective ratio of the selected quadruple can be determined according to the ratio of points on the convex hull polygon of the character to be recognized transformed into the perspective invariant coordinate system sex. When the ratio of points on the convex hull polygon transformed into the perspective invariant coordinate system is smaller than a predetermined ratio, it may be determined that the selected quadruple is invalid.

In one example, if the proportion of points on the convex hull polygon of the character to be recognized transformed into the perspective invariant coordinate system is greater than or equal to 90%, the selected quadruple is considered valid.

If the proportion of the points on the convex hull polygon of the character to be recognized transformed into the perspective invariant coordinate system is less than 90%, it can be determined that the selected quadruple is invalid, discard the quadruple, and re-select.

Those skilled in the art can understand that the aforementioned predetermined ratio can also be other values than 90%.

Fig. 3B shows the situation that the point on the convex hull polygon of the character to be recognized is transformed into the situation that the ratio under the perspective invariant coordinate system is less than 90%. In this case, the selected quadruple group is invalid. be discarded in the process.

Next, in step S106, features are extracted from the converted character to be recognized to obtain a feature vector of the character to be recognized.

Preferably, the perspective-invariant coordinate system may be divided into multiple sub-regions, a histogram is constructed according to the number of pixels of the character to be recognized in each sub-region, and the histogram is used as a feature vector. For example, the perspective-invariant coordinate system may be divided into m×m squares, where m is an integer greater than 1. In this way, a histogram in which the abscissa is the serial number of the sub-region and the ordinate is the number of character pixels contained in the sub-region can be obtained, and the histogram is used as the feature vector fcurrent. Further preferably, the histogram can be normalized, so that the normalized histogram can be used as the feature vector fcurrent. Those skilled in the art can understand that features such as edge direction and gradient information can also be extracted to form feature vectors.

In step S108, a record matching the obtained feature vector of the character to be recognized is searched in the pre-stored character category table, and a vote is performed for the character category corresponding to the found record.

Specifically, each record in the character category table contains a character category and a corresponding feature vector. Wherein, the feature vector may be in the form of a histogram, a normalized histogram, or the like.

In an example, the character category table may be a hash table, wherein each record in the hash table further includes an index value obtained by hashing the feature vector therein.

In this example, hash processing is performed on the feature vector, for example, uniform binary quantization is performed, and the quantized feature vector is converted into an index value bin. Those skilled in the art can understand that multi-level quantization such as three-level quantization can also be performed on the feature vectors, and the quantized feature vectors can be converted into index value bins.

According to the obtained bin value, the corresponding character category d and feature vector fstored are found in a pre-stored character category table (such as a hash table).

Compare the Euclidean distance between the feature vector fcurrent of the character to be recognized and the feature vector fstored stored in the character category table, if ‖fcurrent-fstored‖ is smaller than a small predetermined value, vote for the character category d.

If there are multiple votes for the same character category d for the same bin value, the other votes for the same character category d from the bin value are ignored, ie, only one vote is counted.

FIG. 4A shows that the feature vector is obtained by dividing the perspective invariant coordinate system into 4*4 squares as an example. FIG. 4B shows a histogram obtained by normalizing the number of character pixels in each square as a feature vector. The eigenvectors are uniformly binary quantized, and the quantized eigenvectors are converted into index value bins, that is, 40944 in the hash table shown in FIG. 4C . According to this value, the character category d and the corresponding feature vector in the hash table can be found.

For the character category table in step S108, the above-mentioned character category table can be constructed in the following manner for each template character used for learning in the learning phase before character recognition: first, the feature vector of the template character will be extracted, including: using Cross-ratio selects a quadruple of ordered four points from the convex hull polygon of the template character; transforms the template character into the perspective invariant coordinate system determined by the selected quadruple; from the transformed template character Extract features to obtain the feature vector of the template character. The steps of extracting the feature vectors of the template characters are similar to the steps of obtaining the feature vectors of the characters to be recognized above, and will not be repeated here.

It should be noted that after the template character is transformed into the perspective-invariant coordinate system, the effective value of the selected quadruple should be determined according to the ratio of points on the convex hull polygon of the template character transformed into the perspective-invariant coordinate system. sex. If the selected quadruple is invalid, discard the quadruple and re-select. The method for determining the validity of the selected quadruple here is similar to the above method for determining the validity of the selected quadruple for the character to be recognized, and will not be repeated here.

After the feature vector of the template character is extracted, the character category of the template character and the obtained feature vector are stored as a record in the character category table. Then, repeat the multiple steps of selecting ordered quadruples of four points from the convex hull polygon of the template character until a record in the character category table is obtained, until all quadruples on the convex hull polygon of the template character are traversed Group. Thus, a character category table including each template character is constructed.

It should be noted here that although the above describes how to generate the character category table, those skilled in the art should understand that for the method for recognizing characters according to the embodiment of the present invention, it is only necessary to pre-store one of the above The above character category table is sufficient, and there is no need to care about how the character category table is generated.

Next, in step S110, the above steps are repeated for a predetermined number of times for different quadruples on the convex hull polygon of the character to be recognized. Here, the predetermined number of times can be a preset number of times, or it can be selected every other point or every two points among the points on the convex hull polygon to perform the above steps S102-S108, that is, to complete a vote .

Finally, in step S112, the character category with the most votes is determined as the character recognition result.

The above character categories can also be used in a broad sense. For example, in the case of multiple fonts, the fonts A-Arial and A-Calibri can be counted as one type, or can be counted as two types to distinguish and recognize.

According to the character recognition method of the present invention, character recognition can be performed even when the characters under the large perspective deformation shown in FIG. 5A and the characters shown in FIGS. 5B and 5C are damaged or missing.

Fig. 6 shows a block diagram of an apparatus for character recognition according to an embodiment of the present invention. The character recognition device 600 includes a selection unit 602 , a transformation unit 604 , an extraction unit 606 , a voting unit 608 , a repetition control unit 610 and a determination unit 612 . The selection unit 602 is configured to select an orderly quadruple composed of four points from the convex hull polygon of the character to be recognized by using the cross ratio. The transformation unit 604 is configured to transform the character to be recognized into a perspective-invariant coordinate system determined by the selected quadruple. The extraction unit 606 is configured to extract features from the transformed character to be recognized to obtain a feature vector of the character to be recognized. The voting unit 608 is configured to search a pre-stored character category table for a record matching the obtained feature vector of the character to be recognized, and vote for the character category corresponding to the found record. The repetition control unit 610 is configured to repeat the above steps for a predetermined number of times for different quadruples on the convex hull polygon of the character to be recognized. The determining unit 612 is configured to determine the character category with the most votes as the character recognition result.

Similarly, the character category table may be pre-stored in the device for character recognition, or a character category table may be constructed for each template character used for learning in the learning phase before character recognition. The manner of constructing the character category table here is similar to the manner of constructing the character category table in the embodiment of the character recognition method of the present invention described above with reference to FIGS. 1 to 4 , and will not be repeated here.

Optionally, the character recognition device 600 further includes: a judging unit (not shown), configured to determine whether a point on the convex hull polygon of the character to be recognized is transformed into a perspective-invariant coordinate system according to a ratio smaller than a predetermined ratio. It is judged whether the selected quadruple is invalid, and if it is determined that the selected quadruple is invalid, the quadruple is discarded.

Optionally, the character category table is a hash table, where each record in the hash table also includes an index value obtained by hashing the feature vector therein.

Optionally, the extraction unit 606 includes a segmentation subunit 6062 and a histogram construction subunit 6064, the segmentation subunit 6062 is configured to divide the perspective invariant coordinate system into multiple sub-regions, and the histogram construction subunit 6064 is configured to A histogram is constructed according to the number of pixels of the character to be recognized in each sub-region, and the histogram is used as a feature vector.

Details about the operation and functions of each part of the character recognition device 600 can refer to the embodiment of the present invention described in conjunction with FIG. 1 to FIG. 4 , and will not be described in detail here.

It should be noted here that the structure of the character recognition device 600 and its constituent units shown in FIG. 6 and FIG. 7 are only exemplary, and those skilled in the art can make detailed descriptions of the structural block diagrams shown in FIG. 6 and FIG. 7 as needed. Revise.

The basic principles of the present invention have been described above in conjunction with specific embodiments, but it should be pointed out that those skilled in the art can understand that all or any steps or components of the method and device of the present invention can be implemented on any computing device (including processors, storage media, etc.) or a network of computing devices, implemented with hardware, firmware, software, or a combination thereof, this is a person of ordinary skill in the art who uses their basic knowledge after reading the description of the present invention programming skills will do.

Therefore, the object of the present invention can also be achieved by running a program or a group of programs on any computing device. The computing device may be a known general-purpose device. Therefore, the object of the present invention can also be achieved only by providing a program product including program codes for realizing the method or device. That is, such a program product also constitutes the present invention, and a storage medium storing such a program product also constitutes the present invention. Obviously, the storage medium may be any known storage medium or any storage medium developed in the future.

In the case of implementing the embodiments of the present invention by software and/or firmware, the program constituting the software is installed from a storage medium or network to a computer having a dedicated hardware configuration, such as a general-purpose computer 800 shown in FIG. When there are various programs, it is possible to perform various functions and so on.

Fig. 8 shows a schematic block diagram of a computer that can be used to implement the method and device according to the embodiments of the present invention. In FIG. 8 , a central processing unit (CPU) 801 executes various processes according to programs stored in a read only memory (ROM) 802 or loaded from a storage section 808 to a random access memory (RAM) 803 . In the RAM 803 , data required when the CPU 801 executes various processes and the like is also stored as necessary. The CPU 801 , ROM 802 , and RAM 803 are connected to each other via a bus 804 . The input/output interface 805 is also connected to the bus 804 .

The following components are connected to the input/output interface 805: an input section 806 (including a keyboard, a mouse, etc.), an output section 807 (including a display such as a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.), A storage section 808 (including a hard disk, etc.), a communication section 809 (including a network interface card such as a LAN card, a modem, etc.). The communication section 809 performs communication processing via a network such as the Internet. A driver 810 may also be connected to the input/output interface 805 as needed. A removable medium 811 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like can be mounted on the drive 810 as needed, so that a computer program read therefrom can be installed into the storage section 808 as needed.

In the case of realizing the above-described series of processing by software, the programs constituting the software are installed from a network such as the Internet or a storage medium such as the removable medium 811 .

Those skilled in the art should understand that such a storage medium is not limited to the removable medium 811 shown in FIG. 8 in which the program is stored and distributed separately from the device to provide the program to the user. Examples of the removable media 811 include magnetic disks (including floppy disks (registered trademark)), optical disks (including compact disk read only memory (CD-ROM) and digital versatile disks (DVD)), magneto-optical disks (including )) and semiconductor memory. Alternatively, the storage medium may be the ROM 802, a hard disk contained in the storage section 808, or the like, in which the programs are stored and distributed to users together with devices containing them.

The invention also proposes a program product storing machine-readable instruction codes. When the instruction code is read and executed by a machine, the above-mentioned method according to the embodiment of the present invention can be executed.

Correspondingly, a storage medium for carrying the program product storing the above-mentioned machine-readable instruction codes is also included in the disclosure of the present invention. The storage medium includes, but is not limited to, a floppy disk, an optical disk, a magneto-optical disk, a memory card, a memory stick, and the like.

In the above description of specific embodiments of the present invention, features described and/or illustrated for one embodiment can be used in the same or similar manner in one or more other embodiments, and features in other embodiments Combination or replacement of features in other embodiments.

It should be emphasized that the term "comprising/comprising" when used herein refers to the presence of a feature, element, step or component, but does not exclude the presence or addition of one or more other features, elements, steps or components.

In addition, the method of the present invention is not limited to being executed in the chronological order described in the specification, and may also be executed in other chronological order, in parallel or independently. Therefore, the execution order of the methods described in this specification does not limit the technical scope of the present invention.

Although the present invention has been disclosed by the description of specific embodiments of the present invention above, it should be understood that all the above embodiments and examples are illustrative rather than restrictive. Those skilled in the art can devise various modifications, improvements or equivalents to the present invention within the spirit and scope of the appended claims. These modifications, improvements or equivalents should also be considered to be included in the protection scope of the present invention.

Claims (8)

1. a character identifying method, including:

Double ratio is utilized to select the four-tuple that orderly four point is constituted from the convex closure polygon of character to be identified；

By in described character transformation to be identified to the perspective invariant geomagnetic coordinates determined by selected four-tuple；

The ratio that the point on convex closure polygon according to described character to be identified is converted under described perspective invariant geomagnetic coordinates is No judge that selected four-tuple is the most invalid less than predetermined ratio, if it is determined that selected four-tuple is invalid, then give up This four-tuple, re-executes and utilizes double ratio to select the quaternary that orderly four point is constituted from the convex closure polygon of character to be identified The step of group；

From the character described to be identified after conversion, extract feature obtain the characteristic vector of character to be identified；

The record mated with the characteristic vector of the character to be identified obtained, Xiang Cha is searched in the character class table prestored The character class corresponding to record found is voted；

With pre-determined number, above-mentioned steps is repeated for the different four-tuple on the convex closure polygon of described character to be identified；With And

The character class obtaining ballot poll most is defined as character identification result.

2. the method for claim 1, each template character for study is passed through by wherein said character class indicator In the following manner builds:

Double ratio is utilized to select the four-tuple that orderly four point is constituted from the convex closure polygon of template character；

Described template character is transformed in the perspective invariant geomagnetic coordinates determined by selected four-tuple；

From the described template character after conversion, extract feature obtain the characteristic vector of described template character；

The characteristic vector of the character class of described template character with the described template character obtained is deposited as a record In described character class table；And

Repeat the above until all four-tuple on the convex closure polygon traveling through described template character, in order to build described word Symbol category table.

3. method as claimed in claim 1 or 2, wherein said character class table uses Hash table, in wherein said Hash table Every record also comprise index value characteristic vector therein being carried out Hash process and obtain.

4. the method for claim 1, wherein extracts feature from the character described to be identified after conversion and obtains and wait to know The step of the characteristic vector of malapropism symbol includes: described perspective invariant geomagnetic coordinates is divided into many sub regions, knows according to described waiting Malapropism symbol number of pixels in each sub regions builds rectangular histogram, using described rectangular histogram as described characteristic vector.

5. a character recognition device, including:

Select unit, be arranged to utilize double ratio to select orderly four point to constitute from the convex closure polygon of character to be identified Four-tuple；

Converter unit, is arranged to described character transformation to be identified to the constant seat of the perspective determined by selected four-tuple In mark system；

Judging unit, the point being arranged on the convex closure polygon according to described character to be identified is converted to described perspective not Become less than predetermined ratio, whether the ratio under coordinate system judges that selected four-tuple is the most invalid, if it is determined that selected Four-tuple is invalid, then give up this four-tuple；

Extraction unit, is arranged to extract feature from the character described to be identified after conversion to obtain the spy of character to be identified Levy vector；

Ballot unit, is arranged in the character class table prestored search and the feature of the character to be identified obtained The record of Vectors matching, votes to the character class corresponding to the record found；

Repetitive controller unit, the different four-tuple being arranged on the convex closure polygon for described character to be identified are with predetermined Number of times repeats the operation that above-mentioned unit is carried out；And

Determine unit, be arranged to the character class obtaining ballot poll most is defined as character identification result.

6. device as claimed in claim 5, wherein, each template character for study is passed through by described character class indicator In the following manner builds:

Double ratio is utilized to select the four-tuple that orderly four point is constituted from the convex closure polygon of template character；

Described template character is transformed in the perspective invariant geomagnetic coordinates determined by selected four-tuple；

From the described template character after conversion, extract feature obtain the characteristic vector of described template character；

The characteristic vector of the character class of described template character with the described template character obtained is deposited as a record In described character class table；And

Repeat the above until all four-tuple on the convex closure polygon traveling through described template character, in order to build described word Symbol category table.

7. the device as described in claim 5 or 6, wherein said character class table uses Hash table, in wherein said Hash table Every record also comprise index value characteristic vector therein being carried out Hash process and obtain.

8. device as claimed in claim 5, wherein said extraction unit includes splitting subelement and rectangular histogram builds subelement, Described segmentation subelement is arranged to be divided into described perspective invariant geomagnetic coordinates many sub regions, and described rectangular histogram builds son Unit is arranged to build rectangular histogram, described Nogata according to the described character to be identified number of pixels in each sub regions Figure is used as described characteristic vector.