JPH0245892A - Method and device for recognizing character - Google Patents

Abstract

PURPOSE:To execute plural threshold deciding methods by executing plural threshold decision processings in parallel, executing the processing of a preprocessing course in parallel and judging an input character from plural results identified through a feature extracting course and an identifying course. CONSTITUTION:By a first preprocessing part 5, a threshold is decided in, for example, a mean linear expression method, and image data are binarized by the threshold. By a second preprocessing part 6, the threshold is decided in, for example, a p-Tile method, and the image data are binarized. By a third preprocessing part 7, the threshold is decided in, for example, a discrimination analysis method-Ohtsu, and the image data are binarized. By a feature extracting part 3, feature quantities are calculated to three binarized data to be normalized, respectively, they are sent to an identifying part 4, and by comparing the quantities with the feature quantity of a dictionary inside a computer, what character the input character is is identified, and three candidate characters to respective threshold decision methods are decided.

Description

[Detailed description of the invention] [Industrial application field] This invention is in the preprocessing process of nine character recognition.

A character recognition method and device capable of supporting character recognition under all conditions by executing a plurality of threshold determination methods having various advantages and processing using the obtained thresholds. 1.

[Conventional technology]

In general, in the preprocessing process in the nine character recognition process.

Processes such as removing noise and distortion components contained in the image data and converting the position and size of characters are performed to facilitate the feature extraction and recognition process that follows the preprocessing process.

Binarization of image data by threshold processing is known as the most common method of noise removal in the preprocessing process. In other words, for the 9 images IP (i, j) with respect to the threshold t, the density of each pixel is JP (i, j) = 1: IP (i, j) ≧ t
O: Processes IP (i, j) < t.

There are various methods for determining this threshold, each with different advantages.

An image that has been binarized using the threshold set in this way is sent to the feature extraction/identification process after passing through the lf processing process, so if it is binarized using an inappropriate threshold, it will cause errors. It may be recognized appropriately.

When performing binarization in two-character recognition in this way, an important issue is how to set the optimal threshold for one image under all conditions.

One conventional example of a threshold value determination method is a method called the average-order method, in which a threshold value is obtained from the average density of each pixel. This method has the advantage that a threshold value can be determined without determining the density histograms of two images, and that it requires a very simple algorithm and requires a small number of steps.

Regarding the conventional example of character recognition method using mean-order method,
Third cause.

This will be explained using FIG.

FIG. 3 shows a functional block diagram of a conventional embodiment, and FIG. 4 shows an example of a preprocessing section of the conventional embodiment.

In the figure, 111 is a scanning section, (2) is a preprocessing section, and (3
) is a feature extraction unit, (4) is an identification unit, and (8) is a threshold value determination process.

(9) is a binarization process, and +IG is a normalization process. A scanning unit 1 that illuminates input characters and obtains shading information of the input characters! ) image data contains various noises, distortions, etc., so it is difficult to identify them as is. This data is then sent to the preprocessing section (2). In the preprocessing section (2), a threshold value is set using the mean-order equation method in the threshold value determination process (8), and the data is divided into two according to the threshold value.
Binarization processing (19+), which makes recognition easier (
In order to do this, normalization processing + IG is performed to enlarge or reduce characters of various sizes to a fixed size to match the dimensions.

Next, the characters are sent to the feature extraction unit (3), which determines the characteristics of the characters necessary to recognize the nine characters, and then matches the characters with a set of dictionaries that serve as a reference in the character recognition device. .

The information is sent to an identification unit (4) that performs recognition.

Conventional character recognition methods limit the threshold value determination method to one method, the average-order equation method, as described above, determine the threshold value, and use that threshold value to perform binarization and identification. I was going.

[Problem to be solved by the invention]

In the conventional character recognition method, the threshold value determination method is limited to one method, the mean-order equation method, as described above. If the input character conditions (ink bleeding, blurring, etc. of printed characters) are constant, there is no problem with the conventional method, but such cases are rare, and in reality, each input character The conditions are different.

In this case, the threshold value determination method called the limited mean-order equation method is not necessarily optimal for each condition. For example, the P-tile method is considered to be optimal when the ratio is stable, such as for typefaces, lines, symbols, etc. In addition, automatic vehicle number reading systems are subject to brightness and darkness depending on the weather.

Various conditions can be considered, such as blurring of numbers due to speed, and in such cases, a flexible threshold determination method such as the 9-discriminant analysis method--Otsu's method seems to be optimal.

In other words, characters that could not be recognized by the threshold set by the limited threshold determination method.

There is a possibility that it can be recognized with a threshold value set by another threshold determination method.

This invention was made to solve these problems, and it is a character recognition system that emphasizes the setting of a threshold value in single character recognition and enables the execution of multiple threshold value determination methods with various advantages. The object of the present invention is to provide a method and an apparatus thereof.

[Means to solve the problem]

The present invention executes a plurality of threshold determination processes in parallel in the lif processing unit (2) in single character recognition, thereby executing the surface processing process subsequent to the threshold determination process in parallel, The input character is determined from a plurality of results identified through a feature extraction process and a recognition process.

[For production]

In the present invention, when performing threshold processing in single character recognition, the recognition rate for two characters is increased by not limiting the threshold value determination method to one.

〔Example〕

Threshold determination methods include mean-order method, P-tile method,
There are modes such as the mode method 2 discriminant analysis method - Otsu's method, etc., and each has different characteristics.

The mean-order equation method is as explained in section 9.

The P-tile method is a method that determines the threshold value based on the ratio of the area of the a-degree histogram, and the mode method is a method that finds a valley between the peaks (modes) in the density histogram and calculates the density value of that valley. 9 Discriminant analysis method - Otsu's method is a method that uses the criterion of discriminant analysis to maximize the class separation of the concentration histogram.If the histogram is bimodal, it is used as a mode method. This is a great way to automatically determine the threshold even if there is no mode.

Hereinafter, the present invention will be described with reference to nine drawings, taking as an example a case where three threshold value determination methods are executed from among these threshold value determination methods.

FIG. 1 is a block diagram of an embodiment of the invention, and FIG. 2 is a flowchart of an embodiment of the invention.

The given input character is converted into image data at the scanning section (lth step) (step Qυ).The image data that has passed through the scanning section (1) is then sent to the preprocessing section (2), which undergoes three steps. In order to execute the threshold value determination process, the data is passed through the first preprocessing unit (5), the second preprocessing unit (6), and the third preprocessing unit (7).
).

In the preprocessing section, steps 4 to @ are executed.

The first preprocessing unit (5) determines a threshold value by, for example, the mean-order method (step 23), and binarizes the image data using the threshold value (step 0).

Further, in order to facilitate recognition, characters of various sizes are enlarged or reduced to a certain size and normalized by matching the dimensions (step Q4).

The second preprocessing unit (step cl!3 in which the threshold value is determined by a method 2 different from the threshold value determination method of the first preprocessing unit, for example, the P-tile method), the threshold value The image data is binarized (step c13).

Furthermore, 1ζ normalization is performed in the same way as in the first preprocessing section (step @).

In the third preprocessing unit (7), the threshold value is determined by a method different from the threshold value determination method of the first correction processing unit and the second preprocessing unit 9, for example, the discriminant analysis method - Otsu's method. Step (23): binarize the image data using the threshold value (
Step c13). Further, the first preprocessing unit (51), the second preprocessing unit (6), and the third preprocessing unit perform normalization in the same way as the first preprocessing unit and the second preprocessing unit (step c! 4). All data that has passed through the processing section (7) is sent to the feature extraction section (3).

The feature extraction unit (3) calculates a feature amount for each of the three normalized binary data using a conventional method (for example, calculating a Messi feature amount) (step (c)).

Next, the data is sent to the identification unit (4), which identifies what the input character is by comparing it with the characteristic amount of the character inside the computer, and selects candidates for each threshold value determination method. Decide on three letters.

(Step c!1).

〔Effect of the invention〕

As described above, according to the present invention, by setting multiple thresholds instead of limiting the threshold to one, it is possible to respond to characters under all conditions, thereby improving the recognition rate of nine characters. .

[Brief explanation of the drawing]

Fig. 1 is a block diagram of an embodiment of the present invention, Fig. 2 is a flowchart of an embodiment of the invention, Fig. 3 is a block diagram of a conventional embodiment, and Fig. 4 is a surface processing process of a conventional embodiment. It is a figure showing an example. In the figure, (1: scanning section, (2) surface processing section, (3)
) is a feature extraction unit, and (41 is an identification unit. In FIG. 9, the same or corresponding parts are designated by the same reference numerals.

Claims (2)

[Claims] (1) a scanning process for obtaining shading information of a given input character by illuminating the input character; a preprocessing process for removing noise and distortion components contained in the image data obtained in the scanning process; A feature extraction process that extracts character features necessary for reading characters from the image data that has undergone the preprocessing process, and a process that determines what the input character is based on the features obtained in the feature extraction process. In the character recognition process, which consists of the identification process of determining Character recognition characterized in that the image data that has undergone the scanning process is subjected to the plurality of threshold determination processes and the preprocessing process that follows the plurality of threshold determination processes in parallel. Method. (2) a scanning unit that obtains shading information of the input character by illuminating the input character; a preprocessing unit that removes noise and distortion components contained in the image data obtained by the scanning unit; A feature extraction unit that extracts character features necessary for reading characters from the image data that has passed through the preprocessing unit, and a feature extraction unit that determines what the input character is based on the features obtained by the feature extraction unit. The character recognition device includes a first preprocessing unit that determines a threshold value using a first threshold determination method, binarizes and normalizes data, and a second preprocessing unit that determines a threshold value using a first threshold determination method. a second preprocessing unit that determines a threshold using a threshold value determination method and binarizes and normalizes the data;
A character recognition device characterized by comprising a preprocessing unit including a third preprocessing unit that determines a threshold value using a third threshold value determination method and binarizes and normalizes data.