JPS63108480A - System for reading two-dimensional image - Google Patents

Abstract

PURPOSE:To easily extract a characteristic and to attain a correct recognition even at the time of the two-dimensional image such as a low quality character by dividing a circumscribed frame, integrating all direction codes of a divided area and detecting a maximum value. CONSTITUTION:When a two-dimensional image such as a character is read, first, a circumscribed frame 2 is divided into 9 pieces and divided areas A1-A9 are formed. Next, from this, respective areas A1-A9 are obtained by integrating the number of the same direction code di. Namely, for example, all direction codes in the divided area A1 are integrated, the maximum value of the integrating value is detected and it can be decided which direction code is predominant at every area. The direction code and the number, which go to be the maximum value, are made into the feature value to show the partial feature of a character 1A at the divided area corresponding to the vector quantity having a constant direction and a length. Consequently, by obtaining all feature values of respective areas A1-A9, the feature can be easily extracted and correctly recognized even at the time of the two-dimensional image such as a low quality character.

Description

[Detailed Description of the Invention] (Industrial Application Field) The present invention extracts the features of a two-dimensional image such as characters or figures by tracking the boundary points between the pixels constituting the image and its background. The present invention relates to a two-dimensional image reading method for recognizing two-dimensional images.

(Prior Art) FIG. 8 shows a conventional reading device used for this type of application, in which 31 is a two-dimensional imaging device such as a camera using COD, and 32 is a two-dimensional imaging device 31. A/ that converts analog data into digital image data
A D converter, 33 is a frame memory for storing digital image data, 34 is an arithmetic unit such as a microprocessor that performs processing such as binarization, feature extraction, and determination of the digital image data, and 35 is an arithmetic unit that stores processing results by the arithmetic unit 34. It is an input/output device for outputting data to the outside.

This reading device scans, for example, binarized image data in a fixed direction and stores the address of the pixel that is the point of change from logic "0" to logic "1" or from logic "1" to logic "0". By doing so, features are extracted, and after the completion of all scanning, the two-dimensional image reproduced based on the address is compared with a standard pattern to recognize or judge the image.

(Problems to be Solved by the Invention) However, when reading a two-dimensional image such as a character using such a reading device, conventionally there have been the following problems. In other words, there are various types of characters that can be read, such as those engraved with wire dot printers and those that are recently recorded using lasers. Due to shortages, letters may be faded or partially missing.

Or there were cases where it was unclear. Furthermore, due to scratches, roughness, stains, etc. on the background of paper on which characters are recorded, good quality data may not be obtained even if digital image data is fixed or floating binarized.

Furthermore, even assuming that high-quality binary image data can be obtained, it is extremely difficult to accurately recognize two-dimensional images such as faded or missing low-quality characters using the conventional reading methods described above. .

The present invention was proposed to solve the above problems, and its purpose is to easily extract the characteristics of two-dimensional images such as low-quality characters so that they can be recognized accurately. The object of the present invention is to provide a method for reading two-dimensional images.

(Means for Solving the Problems) In order to achieve the above object, the present invention extracts features of two-dimensional image data such as binarized characters and figures, and compares them with a standard pattern to create a two-dimensional image. In a two-dimensional image reading method that recognizes By tracking the boundary points between the pixels that make up the two-dimensional image and their background, for example, eight types of direction codes with a certain direction are associated with each pixel, and one of the plurality of direction codes is selected for each divided area. The present invention is characterized in that the number of codes in the same direction is accumulated, a feature value is obtained from the direction code that has the maximum value, and the number thereof, and the two-dimensional image is recognized based on this feature value.

It is also effective, if necessary, to group a plurality of adjacent divided regions into one group, integrate the number of codes in the same direction within this group, and obtain the feature value from the maximum value.

(Operation) By tracking the boundary point between one pixel and the background within the divided area, the pixel is associated with a direction code having a certain direction. This process is performed for all divided areas to store each direction code of the boundary point pixel, and then
The number of the same direction code, in other words, the number of pixels having the same direction code, is added up for each divided area, and the maximum value of the direction code and its number are used as feature values for feature extraction. Note that the feature values may be obtained by grouping a plurality of adjacent divided regions into one group.

This feature value is a vector quantity consisting of a direction code with a fixed direction and its number, and corresponds to the partial direction and length of the two-dimensional image to be read. By detecting all feature values, the overall features of the two-dimensional image to be compared with the standard pattern can be extracted and recognized.

(Example) An embodiment of the present invention will be described below with reference to the drawings.

First, Figure 1 shows a character as a two-dimensional image to be read, and this character 1 is OCR-A (JISC62
50) is the number "1" represented by the font.

In the present invention, when reading this character 1, a circumscribing frame 2 in which the character 1 circumscribes is provided as shown in the figure, and the boundary points between the pixels within this circumscribing frame 2 and the background of paper etc. are tracked, and the The direction is defined counterclockwise with the character 1' as the center, as shown by the arrow in FIG. The starting point for tracking is the minimum X when assuming a coordinate axis consisting of an X axis and a Y axis around the character 1'' as shown in FIG.

Let it be a point of a pixel that has a Y coordinate point. That is, the third
In the illustrated example, 1a is the tracking start point. Note that the tracking direction and the tracking start point of the boundary point are not limited to this example at all, and can be set freely.

Next, as shown in FIG. 4, eight types of direction codes are set, from do to d7, indicating in what direction the pixel having the boundary point is located. This direction code indicates in which direction the adjacent pixels forming the boundary point are located with respect to the currently focused pixel when the boundary point is tracked according to the above-mentioned tracking direction.For example, the direction code d0 means that the adjacent pixel is on the right side in the horizontal direction. It should be noted that these direction codes can be divided into further subdivided directions to provide even more types.

Under the above premise, a preprocessing method for recognizing the image after A/D converting and binarizing a two-dimensional image input by a two-dimensional imaging device will be described below.

That is, a boundary point file as shown in Table 1 is created for each pixel constituting the boundary point between the character and the background as a two-dimensional image. This boundary point file consists of the X and Y coordinate values of a pixel and the direction code d1 of that pixel.

Table 1 Next, circumscribing frame 2 is created according to the shape characteristics such as the font of one character.
Divide into multiple parts. The number of divisions is arbitrary, and a larger number is preferable in order to improve reading accuracy, but in this embodiment, as shown in FIG. 5, the circumscribing frame 2 is divided into nine parts to form divided areas A1 to A. do. In addition, in Figure 5, IA
As in Figure 1, OCR-A (JIS C6250)
Indicates the characters expressed by the font, and x1 to x4
．． y1 to y4 indicate X and Y coordinates, respectively.

Then, for each of these divided areas A□ to A, the number of pixels with the same direction code di, in other words, the same direction code d in Table 1, is accumulated and calculated 1. For example, divided area A L (The X and Y coordinates of this area are x1≦X≦x
,, y1≦Y≦y2), the number of pixels of the direction code d0, the number of pixels of the direction code d, etc.
Integrate for all direction codes up to 7. This calculation is performed in divided area A9 (X and Y coordinates of this area are X, ≦X≦x
4. y3≦Y≦y4), and by detecting the maximum value of the integrated value, it is possible to determine which direction code is dominant for each region.

Since the direction code with the maximum value and its number can be said to be a vector quantity with a fixed direction and length, this vector quantity can be used as a partial characteristic of the character IA in the divided area. It can be a feature value indicating. Therefore, by determining all the feature values of each of the divided regions A to A9, the overall features of the two-dimensional image to be compared with the standard pattern can be extracted and recognized.

Furthermore, by focusing on the continuity of each part constituting a character, feature values may be determined not for each individual divided region but for a series of a plurality of divided regions.

In this case, a plurality of divided regions are grouped so that the characteristics such as the font of the characters can be best expressed, and the number of pixels of the same direction code within the group is accumulated. As an example of this grouping, as shown in FIG. 6(a), divided areas A1. A2. A.

Combine them into a group Ad, the same < A4. A, ,AG causes group Ad, ,same<A, ,A, ,A regroove Ad, and as shown in FIG. 6(b), divided area A1. A4. A7 makes the group Ad the same <A,,A
, ,A, causes group Ad, ,also A, ,A, ,A
, to form a group Ad. Then, the number of pixels having the same direction code for each of these groups Ad, .

For example, OCR-A (JIS C625
0) characters represented by the font 1. IB is represented by feature values as vectors as shown in the figure for each group Ad~Ad. In other words, in the example of the same figure (a), in group Ad, the number of pixels applied to the direction code d6 is the maximum value, and the feature value is large and long according to that number)
In addition, in group Ad, the double number related to the direction code d2 is the maximum value, and the feature value has a size and length corresponding to the number. Below, other groups and the same figure (
The same is true for example b).

Therefore, the arithmetic device can recognize what the character is by comparing these feature values with the feature values of a standard pattern previously stored in memory for each character.

In this embodiment, characters have been described as two-dimensional images to be read, but the present invention is of course applicable to reading figures, symbols, and the like. Also.

If you change the number of divided areas and the way they are grouped.

Can read 2D images in a variety of fonts and formats.

(Effects of the Invention) As detailed above, according to the present invention, feature values as vector quantities obtained by integrating direction codes for each portion of binarized two-dimensional image data are extracted and compared with a standard pattern. Therefore, even if some pixels are blurred, chipped, or uneven, or the background is scratched or dirty, the overall feature values are not affected, and two-dimensional images can be recognized easily and accurately. can.

[Brief Description of the Drawings] Figures 1 to 7 show an embodiment of the present invention.
Figure 1 is an illustration of the characters to be read, Figure 2 is an illustration of the boundary point tracking direction, Figure 3 is an illustration of the tracking start point, Figure 4 is an illustration of the direction code, and Figure 5 is an illustration of the divided area. Figures 6 (a) and (b) are explanatory diagrams of divided regions divided into groups, Figures 7 (a) and (b) are explanatory diagrams of feature values corresponding to characters, and Figure 8 1 is a block diagram showing a conventional two-dimensional image reading device. 1, IA, IB, 1', 1"--Character 1a
-Tracking starting point 2... circumscribed frame d0 to d7.
・・Direction code A～A、・・・Divided area Ad1～
Ad,...Group Figure 1 Grass Figure 3 Figure 4 II Figure 6 Figure 6

Claims (1)

[Scope of Claims] A two-dimensional image reading method in which features of binarized two-dimensional image data are extracted and the two-dimensional image is recognized by comparison with a standard pattern, comprising: a circumscription of the two-dimensional image; dividing the frame according to the shape characteristics of the two-dimensional image to form a plurality of divided regions;
Within these divided areas, the boundary points between the pixels constituting the two-dimensional image and their background are tracked, and a direction code having a certain direction is associated with each pixel, and a plurality of The number of the same direction codes among the direction codes is integrated, and the feature value is calculated from the direction code that has the maximum value and its number, and based on this feature value, the above-mentioned 2.
A two-dimensional image reading method characterized by recognizing dimensional images.