JPS61156487A - Picture processing device - Google Patents

Abstract

PURPOSE:To realize a stable pattern which is not affected by a pattern position deviation, a rotation and a size deviation by mounting a means for picture recognition which calculates the direction difference of contour tangent line and codes it to obtain the parameters based on pattern shape. CONSTITUTION:After luster-scanning the picture, the trace starting point of picture element and contour for a pattern is calculated to store its position, and then a adjoining contour point on the right hand side of the contour is calculated to obtain the distance (l) from the starting point. Total periphery length L of contour is obtained by operating the distance (l) from the starting point using a formula, and the contour points of one piece N are extracted based on the L. After the starting point is selected as a extracting point, a contour length (l) from the adjoining contour point and its starting point is sought and the contour points are continuously calculated at every distance l/N until N contour points are extracted. Then the directional difference of contour tangent line direction is sequently calculated to code it. Thus it becomes possible to express the pattern shape features correctly and stably using less informations, thereby realizing high-speed and high pattern recognizing device.

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field] The present invention relates to an image processing device capable of recognizing handwritten characters, printed characters, and all other patterns.

[Prior art]

There have been various proposals regarding pattern recognition. For example, when performing pattern recognition using contour information of a pattern, the contour line directions at each contour point are 1 to 8 shown in FIG.
It was sought after as a childish coat according to the directional code.

However, in this method, if there is a pattern as shown in FIG. 2, for example, 51 in the figure is taken as the starting point.

The direction code obtained when tracing the contour clockwise is
As shown in Figure 3, the directional code, which originally changes continuously, becomes discontinuous on the code, as shown by parts A and H in Figure t53 (corresponding to a and b in Figure 2). This causes inconvenience during recognition.

This means that changes in the starting point due to rotation of the original pattern, etc., make subsequent recognition difficult, which has the disadvantage of increasing processing time.

〔the purpose〕

In view of the above points, the present invention eliminates the above drawbacks, captures the outline of a pattern as continuous data, and performs image processing that enables stable pattern recognition unaffected by pattern positional shift, rotation, or size change. The goal is to provide equipment.

〔Example〕

Hereinafter, the present invention will be described in detail with reference to the drawings. The image referred to in this application indicates characters, marks, symbols, pictures, etc.

FIGS. 4(a) and 4(b) are diagrams showing an example of an image processing apparatus to which the present invention can be applied. 1 is leak, 2 is OCR
, league or printer, 3 is a printer, and 4 is a computer computer with a display, which can issue control commands for each device and display desired data. 5
is a keychain. Further, in (b), 10 is an image processing apparatus main body of an in-day type, 11 is a light source, and 12 is an original to be read. Note that 13 is an optical system that guides light from the light source 11 and the pattern on the original 12 to a photoelectric conversion element (for example, a planar C0D) 14 (and 15 is a process for processing output data and/or outputting from the photoelectric conversion element 14). It should be noted that the structure is not limited to such a structure, and may also be one of a system equipped with a disk device, a microfilm workstation, etc.

Next, the riij processing section 15 will be explained in detail.

FIG. 5-1 is a block diagram of the processing section 15. 16
is for the input cover turn in the pre-processing circuit.

Performs binarization, noise removal, etc. Next, the CPU 17 and memory 18 extract features of the pattern. Regarding feature extraction, the flow is shown within the broken line in Fig. 5-2. That is, the contour information of the pattern is encoded by performing primary feature extraction.

Depending on the pattern or extraction method, the code is used as is. Furthermore, a feature other than the code string may be extracted by secondary feature extraction. The features obtained above may be checked against a dictionary in the memory 18 or the like, or similarity and correlation may be calculated to determine the pattern. Furthermore, the above feature extraction may be repeated, or other feature extraction may be performed and then the pattern may be determined.

Next, feature extraction according to the present invention will be explained in more detail. In primary feature extraction, a predetermined number of pixels forming the outline of a pattern are first extracted. Next, the directional differences in the contour tangential directions are determined based on the extracted points, and the differences are sequentially coated.

A flowchart of contour point extraction is shown in FIG.

That is, first, the contour tracing starting point is determined by a method such as rask-scanning the image and first determining the pixels that correspond to the pattern, and the position is memorized (STEPI).Next, the adjacent contour point on the contour, for example, on the right, is determined. Search (STEP 3)
), find the distance intersection from the starting point (STEP 4), the distance statement from the starting point is, for example, the following (1), (1')
Accumulate according to the formula.

For example, a poem in which adjacent contour points are located at the top, bottom, right, left = sentence + 1
(1) Poetry in which adjacent contour points are located diagonally = party + 1.4 (1') From the above, the length L of one circumference of the contour can be found up to 5TEP6.

Next, a fixed number N of contour points are extracted based on this value of L. First, select the starting point as the extraction point (STEP 8)
, calculate the long contour sentence from the adjacent contour points and their starting point (S
TEP l l, 12). Here, how to find the intersection follows the previous (1) and (1'). Then, in order to find contour points for each distance #N, the following conditions are determined (STE, PI3).

Sentence≧V on the construction soil (2) Here, N is the predetermined number of contour points to be extracted.If the formula (2) holds true, go back to 5TEP8 and extract the contour points, and For a while, that is, if the next contour point has not been reached, repeat 5TEP11 and 12. Then, when N contour points are extracted, the process ends (
STEPlo).

If the above method is applied to the pattern shown in FIG. 7, for example, the pixels (S+ to 52o) indicated by diagonal lines in the figure are extracted. Although details are omitted, there are N between each contour point. However, N=20, the contour tracing starting point was Sl, and the contour was traced clockwise.

Next, directional differences in contour tangent directions are sequentially determined based on the contour extraction points and coated. For example, when the contour extraction points are lined up as shown in Figure 8, the difference in the contour tangent direction is α in the figure,
For example, as shown in FIG. 9, it is quantized in eight directions and coated.

Here, always make the first contour tangent direction (Si ding completed in FIG. 8) coincide with the direction of the line 1''O.

In this way, by encoding the directional difference in the tangential direction of the contour, a code string can be obtained that does not generate any unreasonable discontinuous points and that changes periodically no matter how many times the code is passed around the contour. - As an example, the code string obtained using all the contour points of the pattern in FIG. 2 is shown in the @IO diagram. As is clear from a comparison with FIG. 3, there are no illegal discontinuities, it is suitable for subsequent identification processing, and is stable against fluctuations in the starting point of the code.

FIG. 11 shows the result of applying the detailed processing to the pattern shown in FIG. 7. Whether it is the coat row shown in parentheses in the figure, the contour shape is well expressed without being affected by the unevenness of the pattern 411, and the length is also normalized.

Next, in secondary feature extraction using the coat string obtained by the primary feature extraction described below, information particularly effective for subsequent identification is extracted from the coat string as feature parameters. An example of the feature parameters is shown below.

Pl : Number of code "4" P2 : Number of code "3" P3 : Number of code "2" P, : Number of code rlJ P5 ・Number of code "0" P6 : Number of code "-1" P7 ・Number Prl of code "-2J", number Pl of code "-3": Number of clusters of negative codes, however, code "0" may be inserted between them.

P+o: Number of clusters of codes less than or equal to r-2J F++:
Number of code clusters of r2J or more F, 2: Number of code clusters of r3J or more P13. Maximum consecutive number of codes "0" P I/, Sum of negative codes In the parameters, for example, Po is equivalent to the number of loops or the number of separated patterns, P9 is equivalent to the number of recesses in the pattern, and both represents the global characteristics of P5. P
l3 represents the linearity of the outline shape of the pattern, and P II
Corresponds to the number of sharp points.

Next, based on the following data, the value of the feature parameter obtained by secondary feature extraction is compared with the dictionary for each category stored in the preliminary memory, and the input pattern is determined. Determine which category it belongs to or which category it belongs to. An example of the dictionary is shown in FIG.

MAX in the diagram.

MIN is the maximum and minimum value allowed for each feature parameter, and the X mark indicates that there is no limit. And P1~P
If all 14 feature parameters or conditions are satisfied, the input pattern is determined to belong to that category.

The determination result may be output as is as a recognition result in 1, or may be stored as a candidate category and proceed to another recognition process.

In the embodiment No. 1111, the contour point extraction method is not limited to the method shown in FIG. 6, but is used to express the characteristics of the pattern shape without being affected by minute irregularities or noise in the contour shape of the pattern. Such a method of extracting a fixed number of pixels can be applied without affecting subsequent processing at all. Furthermore, the direction code is not limited to the eight directions shown in Fig. 9, but can be more detailed (for example, the first
It is effective to roughen the surface (as shown in Figure 3) or to make it rough.

Furthermore, the feature parameters are P0 to F + shown here.
The number is not limited to 4, and various types can be used depending on the recognition target.

In addition to the method of checking against a dictionary as shown in Figure 12, the Kaashi method also calculates the distance #D between the input cover pattern and each category as a function of the feature parameters P0 to Pk, and makes judgments based on the size of D. There is a way to do that.

For example, D=7=a, (b, -P,) 2 (3) 1・
01 However, a, b are constants determined for each category.

〔effect〕

As described in detail above, according to the present invention, a certain number of contour points of a pattern are extracted, a direction difference in the contour tangent direction is determined and coded based on the extracted contour points, and a feature parameter is determined from there.

It is possible to stably and faithfully represent the characteristics of a pattern shape with a small amount of information, and in combination with a simple judgment method, it has become possible to provide a pattern recognition device that is fast and has a high recognition rate as a whole.

Furthermore, a predetermined number of contour points can be extracted using a simple algorithm, and the accuracy of subsequent pattern recognition can be improved.

[Brief explanation of drawings]

Figure 1 is a diagram showing a conventional direction code, Figure 2 is a diagram showing an example of a pattern, Figure 3 is a diagram showing the relationship between contour points and direction codes, and Figure 4 is a diagram to which the present invention can be applied. Figure 5-1 is a block diagram of the processing unit; Figure 5-2 is a diagram showing the flow of pattern feature extraction; Figure 6 is a diagram showing contour points in the primary feature extraction unit. Figure 7 is a diagram showing an example of a pattern and its outline extraction points; Figure 8 is a diagram showing the relationship between outline extraction points and the difference in the direction of the contour tangent; FIG. 10 is a diagram showing the relationship between contour points and direction codes, FIG. 11 is a diagram showing direction codes according to the present invention, and FIG. 12 is a diagram used in the determination unit. A diagram showing an example of a dictionary for each category. FIG. 13 is a diagram showing subdivided direction codes. 16 is a preprocessing circuit, and 1 is a reader. ship +2

Claims (1)

[Scope of Claims] Detection means for tracing contour information of an image pattern in a predetermined direction to detect contour points; Extraction means for extracting a predetermined number of contour points from among the contour points; The direction difference in the contour tangential direction is determined based on the extracted points, and is coded. Parameters that reflect the shape of the pattern are determined, and according to the parameters,
An image processing apparatus comprising an identification means for identifying the image pattern.