JPH0452509B2 - - Google Patents

Description

[Detailed description of the invention]

[Technical Field of the Invention] The present invention relates to a recognition processing device that can easily and effectively modify recognition results obtained by character recognition or monosyllable recognition. [Technical background of the invention and its problems] Recently, with the development of information processing technology, various systems have been developed to recognize and process handwritten characters and monosyllables input by voice, create documents, and file them. has been done. This type of recognition processing device, for example, calculates the feature parameters of an input character or monosyllable, calculates the similarity between a feature pattern made up of these feature parameters, and a standard pattern registered in a dictionary, and calculates the recognition result. It's something you get. However, if there are many individual differences or variable factors in the input data consisting of characters or monosyllables, correct recognition results cannot be obtained reliably, resulting in a reduction in the recognition rate. For this reason, it became necessary to correct the recognition results. Conventionally, for example, in the case of handwritten character recognition, an image of the input character and multiple character candidates obtained by recognition processing of this character are sequentially displayed on the display, and the operator is asked to select the correct character candidate. It is left to the discretion of the public. If the correct character does not exist among the character candidates, the operator directly inputs the correct character by operating a keyboard or the like, thereby correcting the recognition result. However, such correction work places a heavy burden on the operator, and furthermore, the correction work requires a large amount of time, resulting in problems such as poor processing efficiency. On the other hand, recently, attempts have been made to compile the character recognition results obtained through the above-mentioned recognition process into words, and to compare the results with a word dictionary to obtain recognition results at the word level. However, in this word matching, since the plurality of candidate units obtained in the character recognition are not utilized in the matching process, correct word recognition results are often not obtained. In this case, the correction work is carried out by returning to the character recognition level, and in the end,
The correction work was complicated. [Object of the Invention] The present invention has been made in consideration of the above circumstances, and its purpose is to provide a highly practical method that can easily and efficiently correct the recognition results for characters and monosyllables. An object of the present invention is to provide a recognition processing device. [Summary of the Invention] The present invention recognizes input characters or monosyllables, collects the recognition results into words, compares them with a word dictionary, and displays the recognized words on a display. When the displayed recognized word needs to be corrected, multiple word-by-word and character-by-character candidates from which the recognized word was obtained are displayed on the display, and the correct word or character selected from among the displayed candidates is used. The recognition word is modified. [Effects of the Invention] Thus, according to the present invention, word candidates and character candidates obtained in the process of obtaining a recognized word are simultaneously displayed on the display, and a word candidate or a character candidate selected from among them is used to generate a recognized word. The correction work is very easy. Therefore, there is no need to place a large burden on the operator. Moreover, since word candidates and character candidates obtained in the recognition processing process are each effectively used, extremely efficient correction can be performed. Furthermore, when displaying the word or character candidates mentioned above, if each candidate is labeled with alphanumeric characters, etc., it will be possible to easily select the correct candidate, which will have a great practical effect. It can be played. [Embodiment of the Invention] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a schematic configuration diagram of the embodiment device, and the second
The figure is a diagram showing the concept of correction processing in this device. Here, recognition processing of handwritten input characters will be explained as an example. For example, the photoelectric conversion unit 1
It optically reads a word consisting of letters and converts it into an electrical signal. Upon receiving this electrical signal, the preprocessing section 2 cuts out each character and normalizes it to extract its characteristic parameters. ing.
In the case of recognition of monosyllables input as vocalizations, preprocessing is performed by extracting monosyllables from the feature patterns of continuously uttered speech and determining feature parameters for each monosyllable. The character recognition unit 3 calculates the degree of similarity with the standard pattern registered in the character recognition dictionary 4 for each character indicated by the feature parameters, and calculates the degree of similarity between each of the n input characters. We are looking for a maximum of r candidate characters.
In this case, each candidate character is given a score corresponding to its candidate rank, and is used for post-processing, which will be described later. The post-processing unit 5 summarizes a maximum of r character candidates for each of the n characters into words,
This is compared with words registered in advance in the word dictionary 6. Then, for a word that is detected as a match through this matching, a score corresponding to the candidate ranking of the character candidates constituting the word is added to obtain a score. The scores thus obtained for each word are compared with each other, and the word with the highest score is taken as the recognition result for the input character string. This recognized word is displayed on the display section 7.
Note that 8 in the figure is a controller that controls the operations of each of the processes described above, and 9 is a keyboard for inputting various information. However, now, in a device configured in this way, for example, 3 each consisting of "Fukuoka Prefecture" and "Fukuoka City"
Assume that two character words are input and recognized. In this case, character candidates as shown in Table 1, for example, are determined by the character recognition unit 3 from the characteristic pattern of the input character.

[Table] These character candidates are grouped into words and word matching is performed by the post-processing unit 5. As a result, matching results are obtained as shown in Table 2, for example.

[Table] In other words, in this case, the word "Fukuoka Prefecture" is obtained as a recognition result by word matching, and since the result of "no matching word" for the second unit, the character recognition The first-ranked character candidate ``Fuku Douhou'' obtained for each character is obtained as a recognition result, and this is displayed on the display. Incidentally, both of the recognized words obtained in this way are different from the input words and require correction.
On the other hand, in this apparatus, the correction process is performed under the control of the controller 8, as schematically shown in FIG. That is, the word candidates 11 and character candidates 12 obtained in the process of the recognition process described above are extracted, and each of these candidates 11,
Labels 13 consisting of alphanumeric characters, etc. are associated with 12, respectively. These character and word candidates 11 and 12 are displayed on the display 14 together with their labels 13 at the same time. In response, the operator specifies the correction target using a cursor or the like, and then selects and instructs 15 the label of the correct character or word via the keyboard 9. According to the selected information, the displayed recognized word is replaced with a correct word or character, and the correction 16 is performed. This modification process will be explained in more detail using FIGS. 3a to 3g. First, the recognized words obtained by the input word recognition process are displayed on the display as shown in FIG. 3a, for example. Since the displayed recognition word needs to be corrected, the first character of the word to be corrected is designated using the cursor 18, as shown in FIG. 3b. As a result, the word candidates obtained when searching for the above recognized words with labels such as "1", "2", and "3" are also "4", "5", and "6". , "7", "8"
The character candidates obtained for each character of the input word are simultaneously displayed on the display with a label of .
That is, a plurality of word and character candidates obtained in the recognition process that resulted in the recognition word "Fukushima prefecture" for the input word "Fukuoka prefecture" are displayed simultaneously on the display. If there is a correct word among the multiple candidates displayed in this way, that is, in this case, the correct word is "Fukuoka Prefecture" with the label "2" attached, so the operator presses the keyboard 9. Input the label information "2" through the screen. As a result, the recognized word whose first character is the character indicated by the cursor 18 is corrected with the correct word specified by the label, as shown in FIG. 3c. Also, when the cursor is specified as shown in Figure 3d in order to correct the recognition word for the second input word, multiple word and character candidates obtained during the recognition process will be displayed along with their labels in the same way. displayed on the display at the same time. However, in this example, since the corresponding word does not exist as shown in Table 2 above, only character candidates are displayed. Therefore, in this case, correction work is performed character by character using the character candidates. Note that even if there is no matching word among the displayed word candidates, correction is performed on a character-by-character basis using character candidates. However, now the first point specified by this cursor 18 is
Since there is no need to modify the character, the cursor 18 is moved to the next character position as shown in FIG. 3e. Regarding the character "same", since the character "oka" found as the fourth candidate is correct, its label "7" is designated and the character is corrected. This makes the third
As shown in Figure f, the correct character is determined. Thereafter, the cursor 18 is moved to the position of the character ``亢'', and the character ``市'', which was found as the fifth candidate, is designated by its label ``8''. As a result, as shown in FIG. 3g, the recognized word "Fukudouhou" is corrected character by character and converted into the correct word "Fukuoka City". In this case, the controller 8 determines the word to be processed from the information indicated by the cursor 18 for the displayed recognition word, and also determines the character to be processed within the word. When a correction is instructed by the information labeled "1", "2", or "3", this is determined to be a correction based on a word candidate, and the word indicated by the label is used to perform the above processing. Correct the target word. Furthermore, when correction is instructed by the information labeled "4" to "8", the characters to be processed in the word to be processed indicated by the cursor 18 are indicated by the label. The characters that appear will need to be corrected. In this way, correction of recognized words is performed using multiple word and character candidates. In this way, according to this device, the word candidates and character candidates obtained in the recognition processing process are effectively used to correct the recognized words, so the correction efficiency is very high and the correction work is simple. . Therefore, there is no need to place a large burden on the operator. Moreover, since multiple candidates for characters and words are displayed all at once, it is possible to immediately determine whether corrections can be made at the word level or whether corrections at the character level are required. Great practical effects such as optimization can be achieved. Note that the present invention is not limited to the above embodiments. In the embodiment, a case has been described in which words such as prefecture names and city names are recognized and corrected, but data entries such as names and company names may also be used. It goes without saying that recognition processing for speech may also be used. Further, the number and order of word candidates and character candidates to be displayed may be determined according to specifications. Alternatively, the correct character or word may be directly indicated using a cursor, light pen, etc., without using a label. In short, the present invention can be implemented with various modifications without departing from the gist thereof.

[Brief explanation of the drawing]

FIG. 1 is a schematic diagram of an apparatus according to an embodiment of the present invention;
FIG. 2 is a diagram showing the concept of the correction process in the embodiment apparatus, and FIGS. 3a to 3g are diagrams showing an example of the correction process. DESCRIPTION OF SYMBOLS 1... Photoelectric conversion part, 2... Pre-processing part, 3... Character recognition part, 4... Character recognition dictionary, 5... Post-processing part, 6... Word dictionary, 7... Display part, 8... Controller, 9... Keyboard, 11... Word candidates, 12...Character candidates, 13
…label.

Claims (1)

[Claims] 1. A means for recognizing characters input by reading or monosyllables input by speaking in character units or monosyllable units, a means for collating these recognition results into word units and comparing them with a word dictionary, and a recognized word that has been verified. means for simultaneously displaying on a display a plurality of word-by-word and character-by-character candidates from which the recognized word was obtained when the displayed recognized word needs to be corrected; A recognition processing device comprising: means for correcting the recognized word using a correct word or character selected and instructed.