KR940007615B1 - Candidate character classifying method using left and right distance shape in recognizing character - Google Patents

Abstract

The classifing method includes the first step of making an outside frame for a character to isolate; the second step of finding the distance from the left/right side of the frame to the character; the third step of calculating the distance changing ratio from the distances between the neighboring components; the fourth step of deciding the linear section, the increasing/decreasing section, and the jumping section; the final step of classifing the unknown character according to the decision tree.

Description

Candidate Character Classification Using Left and Right Distances in Character Recognition

1 is a diagram showing an example of an 8x8 mesh used in the statistical character recognition method.

2 is a diagram showing a frame circumscribed to a character used in a statistical character recognition method.

3 is a diagram illustrating a method of thinning a character having a thickness when the character is structurally recognized into a character having a thickness of 1;

5 is a flowchart for explaining feature extraction of characters during candidate character classification operation using left and right distance shapes of the present invention.

6 is a view for explaining the distance seen from the left and the right.

7 is a diagram showing an example of section determination according to the present invention.

8 is a view showing a first example of a classification tree according to the present invention.

9 is a block diagram of a candidate character classification apparatus using left and right distance shapes of the present invention.

* Explanation of symbols for main parts of the drawings

11: image scanner 12: interface memory

13: Romans 14: Reference data provider

15: classification data providing unit 16: central processing unit

The present invention relates to a method for classifying candidate characters in character recognition, and in particular, by classifying secondary candidate characters by using a projection profile technique, the primary candidate characters in the shape of the characters viewed from the left and right of the characters. The present invention relates to a candidate character classification method capable of classifying accurate candidate characters and obtaining a high recognition rate.

In general, the character recognition method can be largely classified into two types, one of which is a statistical method of classifying and recognizing characters using the statistical characteristics of the character, and the other of the character recognition methods using the structural characteristics of the characters. There is a structural recognition method.

The former statistical character recognition method mostly uses a mesh as shown in FIG. 1, and uses statistical meshes for all characters by using the distribution of pixels included in each mesh, for example, the number of pixels and the distribution density. The candidate characters are classified according to the obtained statistical values, and the most similar characters are selected and recognized among the characters included in the same classified character group.

The mesh used in the statistical character recognition method creates a box that circumscribes the inputted characters as shown in FIG. 2, and is divided into equally spaced left, right, up, and down directions. It is a collection of small grids.

For each mesh's position (i, j) i = 0, 1, 2, ..., nj = 0, 1, 2, ..., m Get the statistics.

For example, if the mesh mean value of the position (1, 1) is A in all the recognition target characters, and the value of the mesh (1, 1) of the input character is larger than the mean value A of the mesh (1, 1), If the value (1, 1) = 1, the value is small, the specific value (1, 1) = 0 is classified.

If 8 × 8 meshes are defined, the feature for one character is 8 × 8, that is, 64 features are used, and candidate characters can be classified using this feature.

The latter structural recognition method recognizes using the stroke structure of letters. The most widely used method is thinning (thinning or skeletoning) method, and the direction of stroke, size, position, connection between strokes, etc. It is a method of recognizing a character in consideration.

That is, first, the character having a thickness as shown in FIG. 3 is thinned to a character having a thickness of 1, and each of the thinned characters is set to an eight-way code as shown in FIG. The code is defined and extracted, and finally recognized as a character that matches the code.

In the above character recognition method, the statistical character recognition method is relatively well recognized even if the recognition character is complicated and there is noise in the image of the read character, but the recognition rate is significantly reduced as the font is changed.

In addition, it requires a lot of reference files for recognition, so it takes a lot of memory and is likely to be confused when similar characters are entered.

On the other hand, the structural character recognition method recognizes by extracting the strokes that make up the characters, so that a relatively high recognition rate can be obtained regardless of the font, but it is difficult to extract the strokes when the characters are complex, and time is thinning. This had a lot of drawbacks.

Therefore, the object of the present invention is to classify the candidate candidates in the shape of the characters viewed from the left and right of the characters and to classify the candidate candidates using the projection profile technique to classify the candidate candidates and obtain a high recognition rate. In providing a method.

The object of the present invention as described above is to compare the distance between the left and right sides of the step of making a frame to circumscribe the input character, the distance from the left and right sides of the frame to the first encounter with the character, Determining a change amount of the left and right distances; determining a straight line, an increase and a decrease, or a + jump and a-jump section according to the change amount; and extracting a feature of a character based on the determined section; , Which is achieved by forming a classification tree according to the characteristics of the extracted characters to classify the candidate characters. Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

A candidate character classification apparatus using the left and right distance shapes of the present invention is shown. The feature extraction unit divides each section by a left and right distance feature and gives a code suitable for the shape of the section. It consists of a processing module, which is a hardware module based on an algorithm, which is a feature synthesizing unit that determines which character group is combined.

9 is a block diagram of a candidate character classification apparatus using left and right distances of the present invention, and includes an image scanner 11 and an image scanner 11 for scanning an image of an input character. An interface memory 12 for storing image data of a character input through the ROM, a ROM 13 having a built-in algorithm for a candidate character classification method using the left and right distance shapes of the present invention; Perform a reference data providing unit 14 for providing reference data when character features are extracted, and an algorithm stored in the character classification data providing unit 15 and ROM 13 for providing data on character classification during character classification. By taking characters one by one from the interface memory 12, extracting the characteristics of the input characters by comparing them with the reference data of the reference data providing unit 14, and synthesizing the extracted features to provide classification data. It was composed of a central processing unit (16) for classifying the characters from the classification data provided from the 15.

The candidate character classification method using the left and right distance shapes of the present invention consists of extracting the characteristics of the input characters largely and classifying the characters by synthesizing the characteristics of the extracted characters.

The step of extracting the feature of the character comprises the steps of creating a frame that circumscribes the input character, obtaining a distance from the left and right sides of the frame to the first encounter with the character, and obtaining a difference between neighboring distances. Comprising a step of determining the amount of change, and the step of determining the straight line, increase and decrease sections or +,-jump section according to the change amount, the step of classifying the character is classified according to the characteristics of the extracted character, that is, the determined section (Tree) is constituted, and accordingly, the characters are classified.

Among the candidate character classification methods of the present invention, an operation of extracting a feature of a character will be described first.

Characters constituting the character is various, but the features used in the present invention used the left and right distance, in order to determine the section by measuring the left, right distance is obtained in the following order.

(Step 1) First, a frame for circumscribing a character is obtained as shown in the second diagram.

(Step 2) The resulting frame is divided in the transverse direction at equal intervals, and the distance that meets the character for the first time from the left or right in the upper direction to the lower direction of the frame is calculated and stored in the manner as shown in FIG.

At this time, the distance feature seen from the left is called PL [i], the distance feature seen from the right is called PR [i], and the i value corresponds to the height of the character.

That is, if the height of the character is 20 pixels, the distances PL [i] and PR [i] viewed from the left and the right to 20 are calculated, respectively.

(Step 3) After calculating the distances seen from the left and the right, the difference PDL [i] between PL [i] and neighboring PL [i] and the difference PDR [i] between PR [i] and neighboring PR [i] Calculate and record the difference.

That is, (step 3) is a step for determining the amount of change of the value based on the left and right distances obtained in (step 2), and the difference PDL [i] and PDR [i] between neighboring distances respectively indicate the amount of change. It can be expressed as formula (1).

PDL [i] = PL [i]-PL [i-1] ·············· (1)

PDR [i] = PR [i]-PR [i-1]

At this time, if the change amount is "+", it means that the distance is increasing, "-" means that the distance is decreasing, and "0" means that there is no increase or decrease of the distance.

In addition, if the magnitude of the change amount is larger or smaller than a certain value, it means that the distance is jumped by a certain amount or more.

(Step 4) Then, the interval is determined using PDL [i] and PDR [i] extracted in (Step 3).

That is, the increase section, the decrease section, and the straight section are determined, and in particular, when a section in which the increase or decrease is larger than a certain threshold value is determined, the jump section is determined.

Therefore, the features that can be finally extracted can be defined as follows.

① If PDL [i], PDR [i] is less than + threshold, it is determined as minus jump section.

② If PDL [i], PDR [i] is less than -threshold, it is determined as -jump section.

③ If PDL [i] or PDR [i] is greater than 0, it is determined as + increment.

④ If PDL [i], PDR [i] is less than 0, it is determined as -reduction section.

⑤ If PDL [i], PDR [i] is equal to 0, it is determined as a straight line section.

FIG. 7 illustrates an example of section determination in recognition of a candidate character, and shows only section determination for a distance viewed from the left as an example.

Assuming that the height i of the character is 12, the change amount PDL [i] of the distance seen from the left side is obtained by equation (1).

Therefore, after determining the change amount PDL [i] of the distance seen from the left side, the section is determined. In the section from 0 to 1, the section is determined to be a straight section because the change amount is 0, and in the section from 1 to 6, the change amount is determined. Since it becomes +, it is determined to be an increase section, and i is determined to be + jump section because the change amount is greater than + threshold value from 6 to 7, and the range from i to 7 to 8 is changed to more than-threshold value. -It is decided by the jump section, and the interval from i to 8 to 11 is changed to-, so it is determined as the reduction section.

The following is a step of synthesizing the feature of the character following the feature extraction of the character.

(Step 5) When the feature is extracted from the feature extracting unit, the feature must be synthesized to determine which character group has the feature.

That is, since the letter "A" has a reduction (-) characteristic of both the left and the right features, the feature is compressed into a feature of (left) → reduction, (right) → reduction.

Similarly, the letter "B" is defined as (left) → straight, (right) → decrement, increase, decrease, increase.

(Step 6) Accordingly, based on the features extracted earlier, a classification tree is constructed as shown in FIG. 8 to classify the input characters.

According to the present invention as described above, in order to increase the accuracy of character recognition, the character classification method that pre-classifies the character, the conventional recognition method is a statistical recognition method and a structural recognition method by avoiding the alternative of the statistical or structural method. It is characterized by a mixture of recognition methods.

That is, since it is composed based on the shape of the characters seen from the left and right of the characters, there is less influence of misclassification due to the change of the font, and it can be classified at high speed. There is also an advantage that can be used.

Claims (1)

A step for creating a frame circumscribed to the input characters, a step for obtaining a distance from the left and right sides of the frame to the first encounter with the letter, and a change amount between the distance between the neighboring left and the distance on the right Determining a section as a straight section, an increasing and decreasing section, or a + jump and a -jump section according to the obtained change amount, extracting a feature of a character according to the determined section, and extracting a feature of the extracted character. Candidate character recognition method using left and right distance shape during character recognition, comprising the step of classifying candidate characters by constructing a classification tree based on the classification.