JPH05120488A - Line graphic recognizing method - Google Patents

Abstract

PURPOSE:To effectively reduce processing amount related to pre-processing by re-executing recognition processing after executing specific pre-processings for the identical line graphics image data only when recognition processing yields rejection. CONSTITUTION:If the first recognition processing outputs a rejection code, a control part 8 makes the second pre-processing part 4 execute a specified pre-processing. After this processing, a recognition process part 5 performs a recognition processing with the character image data processed with the second pre-processing part 4. Then, the recognizing result is outputted to outside, a processing for one character thus ends. As a result, the execution frequency of a particular pre-processing decreases. In addition, a special processing for deciding necessity of specific pre-processing is not required, so, the entire processing amount is effectively reduced.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method of recognizing characters and other line figures.

[0002]

2. Description of the Related Art Conventionally, a line figure such as a character is recognized by performing a certain pre-processing on the line figure image data and then performing the recognition process. If the recognition result is rejected, it is manually performed. The method of correction is common. The contents of pre-processing vary depending on the recognition target, but filtering (Gaussian filter etc.) for blurring,
There are expansion / contraction processing for correction of faintness, smoothing of line segment boundaries (for example, a preprocessing method for line figure recognition according to the 1989 patent application No. 273339).

FIG. 4 shows a conventional character recognition processing flow. All or part of the data of the original image is input (step 41), and then the character is cut out from the input image data, the character image size is normalized, and other preprocessing is performed (step 42). A feature amount is extracted from the pre-processed character image, a candidate character is obtained using this feature amount (in a narrow sense), recognition processing is executed (steps 43 and 44), and the obtained candidate character is output as a recognition result. (Step 45).

[0004]

As described above, the conventional one-way flow of processing is preprocessing- (recognition processing in a narrow sense) -recognition result output, and preprocessing is always performed.
The method that always performs pre-processing cannot avoid an increase in the amount of processing due to pre-processing, but when the recognition processing method (in a narrow sense) is highly likely to cause misrecognition due to noise (for example, the feature amount of a local area). Based on the feature space)
There is also an unavoidable increase in the amount of processing by pretreatment.

In order to suppress an increase in processing amount due to pretreatment, a method of judging whether pretreatment is necessary and performing pretreatment only when it is necessary is also considered. However, if the number of new processes for determining the necessity of the pre-processing is increased, the processing amount cannot always be reduced.

However, the method of always performing the preprocessing regardless of whether the preprocessing is necessary (for example, whether the image is dirty or not) is not good from the viewpoint of the processing amount.

Therefore, a main object of the present invention is to provide a line figure recognition method capable of effectively reducing the processing amount related to preprocessing.

[0008]

According to the invention of claim 1, the recognition processing is executed for the line graphic image data before the specific preprocessing is performed. If it can be recognized in the first recognition processing, the processing is completed. However, if the result is rejection (rejection), a specific preprocessing is performed on the same line graphic image data, and then the recognition processing is performed again. Is executed and the process is completed.

According to the second aspect of the present invention, the content of the specific preprocessing prior to the second recognition processing is controlled based on the feature amount of the line graphic data obtained in the first recognition processing stage. It

[0010]

When the recognition process uses a method that is less likely to be misidentified due to the influence of noise (for example, a method by stroke analysis), there is a high probability that recognition will be possible even if the recognition process is executed without specific preprocessing. By applying the invention of claim 1, the number of executions of the specific preprocessing is reduced.
Moreover, since no special process for determining the necessity of the specific pre-process is required, the total processing amount can be effectively reduced.

It should be noted that the specific pre-processing mentioned here is a process that can be omitted in the entire pre-processing for recognition, and it is a part of the whole pre-processing, or if all the pre-processing is omitted. If possible, the whole pretreatment is a specific pretreatment. For example, in the preprocessing of character recognition, character cutout,
In addition to the normalization of character image size, various processing is included in relation to the recognition method such as blurring for noise removal, correction of blur / blurring, etc., but character cutting or character cutting and size normalization processing Generally cannot be omitted. Therefore, one or more other processes that can be omitted are selected as specific pre-processes.

Further, the preprocessing for the blur / blurring correction is performed by thinning the line if the image tends to be crushed and conversely thickening the line or filling a small hole if the image is harsh. It is desirable to control the content. In order to control the content of such preprocessing according to the characteristics of the image, generally, a process of scanning the image and estimating the line width from the statistics of the black run length is performed.
The addition of such processing cannot be expected to reduce the overall processing amount.

According to the first and second aspects of the present invention, the first recognition process is executed before the specific pre-process is performed, and the feature amount extracted at that stage is used for the crush / blurring. It may be possible to determine the scale.

For example, according to the feature extraction method for a line figure according to the Japanese Patent Application No. 134879 of 1990, a section which is substantially orthogonal to the direction of a line segment is extracted from a line figure image, and a series of the sections is obtained to obtain a section. Create a cross-section series graph in which areas other than cross-section series are unknown areas (areas of line ends, bends, intersections), cross-section series or unknown areas are represented by nodes, and node connection relationships are represented by arcs. However, if this cross-sectional graph has been created, it is possible to find the average line width fairly accurately and use it as a measure for crushing / blurring.

Therefore, when such a feature extraction method is adopted, according to the invention of claim 2, 1 or 2 is added without adding any special processing for measuring crushing / blurring or measuring the average line width. Since it is possible to appropriately control the content of the specific pre-processing when it is rejected in the recognition process for the second time, it is possible to reduce the processing amount, optimize the pre-processing content, and improve the recognition rate accordingly. ..

[0016]

Embodiments of the present invention will be described below with reference to the drawings.

Embodiment 1 This embodiment is intended for character recognition which is a type of line figure recognition. FIG. 1 is a schematic block diagram showing an example of a device configuration for this embodiment.

In FIG. 1, reference numeral 1 is an image memory for storing binary image data of a document. Reference numeral 2 denotes a first pre-processing unit, which is a process that is not omitted in the pre-processing of character recognition, such as a line cutting process that cuts a line image from document image data, a character cutting process that cuts a character image from a line image, The normalization processing of the size of the cut out character image is executed. The character image data cut out and processed by the first preprocessing unit 2 is stored in the character image memory 3. A second pre-processing unit 4 performs one or more processes (specific pre-processes) that may be omitted from the pre-processes, such as a blurring filter process for removing noise. Reference numeral 5 denotes a recognition processing unit that executes a recognition process for a character image on the character image memory 3 or a character image processed by the second pre-processing unit 4, a feature extraction unit 6 that extracts a feature amount of the character image, and an extraction unit. It is composed of a discriminator that obtains a candidate character by matching the generated feature amount with a recognition dictionary. Depending on the recognition algorithm or the like, the normalization processing of the character image size may be performed in the second preprocessing unit 4 instead of the first preprocessing 2. Reference numeral 8 denotes a control unit that controls each unit.

The processing after the step of inputting the character image data to the character image memory 3 will be described with reference to the processing flow chart of FIG.

The control unit 8 inputs the data of one character image into the character image memory 3 (step 21), and then causes the recognition processing unit 5 to start the processing. In the recognition processing unit 5,
The characteristic extraction unit 6 reads the character image data from the character image memory 3, extracts the characteristic amount thereof, and the discrimination unit 7 obtains a candidate character having certainty or more with respect to the character image data on the basis of the characteristic amount. If such a candidate character is obtained, the code of the candidate character or the like is output as the recognition result, and if the candidate character cannot be obtained, a reject (reject) code is output (steps 22 and 23).

The control unit 8 checks whether or not the recognition result of the recognition processing unit 7 is a refusal code (step 24), and if it is not a refusal code, outputs the candidate character code or the like to the outside (step 25), The process for the character image data is ended. If the recognition result is a rejection code, it is checked whether the second recognition process has been executed (step 26). If it has been executed, the rejection code is output to the outside as a recognition result (step 25), and the process for the character image data is performed. finish.

On the other hand, when the refusal code is output in the first recognition process, the control unit 8 causes the second preprocessing unit 4 to execute a specific preprocessing (step 28). The recognition processing unit 5 is caused to execute recognition processing for the character image data processed by the preprocessing unit 4 (steps 22 and 23), and the recognition result is output to the outside (step 25), and the processing of one character is completed. To do.

Embodiment 2 This embodiment is also intended for character recognition, and its device configuration may be basically the same as that of Embodiment 1, so FIG. 1 will be referred to in the following description. The specific preprocessing executed by the second preprocessing unit 4 includes expansion / contraction processing for correction of crushing / blurring. In addition, the feature extraction unit 6 of the recognition processing unit 5 extracts a feature amount capable of obtaining an average line width quite accurately as in the cross-section series graph. FIG. 3 shows a processing flow after the character image data input step in this embodiment.

The control unit 8 inputs the data of one character image into the character image memory 3 (step 31), and then causes the recognition processing unit 5 to execute the recognition processing (steps 32 and 3).
3). When a valid candidate character is obtained as a recognition result, the control unit 8 outputs it to the outside (step 3
5) The processing of one character is completed.

However, when the result of the first recognition processing is rejection, the control unit 8 obtains the average line width W from the feature amount obtained by the feature extraction unit 6 (step 3
7), threshold values TH1 and TH for judging collapse / blurring
2, TH3 (however, TH1 <TH2 <TH3) is compared with the average line width W to select the content of the expansion / contraction process included in the specific preprocessing of the second preprocessing unit 2 (step 38). ). That is, when W ≦ TH1, the execution of only the expansion process is selected. When TH1 <W ≦ TH2, execution of expansion processing and contraction processing in this order is selected. When TH2 <W ≦ TH3, execution of contraction processing and expansion processing in this order is selected. When TH3 <W, execution of only contraction processing is selected.

Then, the control section 8 causes the second preprocessing section 4 to execute the preprocessing (step 39), and the contents of the expansion / contraction processing in this preprocessing are controlled according to the selection result in step 38. Next, the recognition processing unit 5 is caused to perform recognition processing on the character image data processed by the second pre-processing unit 4 (steps 32 and 33), and the result is output to the outside (step 35) and one character. Ends the process.

Although the embodiment of character recognition has been described above, the present invention can be applied to recognition of line graphics other than characters.

[0028]

As described above in detail with reference to the embodiments, according to the invention of claim 1, it is possible to reduce the processing amount for the preprocessing and achieve the reduction of the processing amount of the entire line figure recognition processing. According to the invention of claim 2, at the same time as reducing the processing amount,
The recognition rate can be improved by optimizing the content of preprocessing.

[Brief description of drawings]

FIG. 1 is a schematic block diagram showing an example of a device configuration for character recognition.

FIG. 2 is a processing flowchart of the first embodiment.

FIG. 3 is a processing flowchart of the third embodiment.

FIG. 4 is a processing flow chart of conventional character recognition.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Image memory 2 1st pre-processing part 3 Character image memory 4 2nd pre-processing part 5 Recognition processing part 6 Feature extraction part 7 Discrimination part 8 Control part

Claims (2)

[Claims] 1. A recognition process is performed on line graphic image data before performing a specific pre-process, and a specific pre-process is performed on the same line graphic image data only when the result is rejection. A line figure recognition method characterized in that the recognition process is executed again. 2. The line figure recognition method according to claim 1, wherein the content of the specific preprocessing is controlled based on the feature amount of the line figure image data obtained in the preceding recognition processing stage. ..