JPS60230278A - Picture processor - Google Patents

Abstract

PURPOSE:To express stably features of a profile of a pattern by obtaining the difference between the profile line direction in each profile point at the time of tracking the profile of the pattern and the profile line direction in the previous profile point and encoding it. CONSTITUTION:A preprocessing circuit 16 executes binalization, removal of noises, etc., with respect to an input pattern from a photoelectric convertor 14. Then a CPU17 and a memory 18 extract features of the pattern. Namely, the CPU17 obtains the difference between the profile line direction in each profile point at the time of tracking the profile of the pattern in a certain direction and the profile line direction in the previous profile point, and encodes it, thereby extracting stably features of the profile of the pattern.

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.
However, if a pattern like the one shown in Figure 2 is rarely found, it can be obtained by starting from 81 in the figure and tracing the contour clockwise. The direction code is as shown in Figure 6, and the direction code that originally changes continuously changes on the code as shown in parts A and H in Figure 6 (corresponding to a and b in Figure 2). This results in discontinuity, which causes inconvenience during recognition.

Furthermore, as a means to prevent the above-mentioned discontinuity, (
), when the direction code changes discontinuously, such as when the direction code changes from "8" to "1", there is a method to maintain continuity by adding or subtracting 8 to the value of the direction code.

According to this method, in the case of the pattern shown in FIG. 2, the code becomes continuous as shown in FIG. 4 (al).

However, with this method, when the pattern contour is circled once and the contour tracking starts, the code at the starting point matches the first code as shown at 81 in Figure 2. do not.

In other words, the code values are discontinuous before and after the starting point. This means that even for the same pattern, the resulting code string will change depending on the position of the starting point. The direction code obtained when this is done is as shown in FIG. 4(b), which is quite different from that shown in FIG. 4(a). This means that changes in the starting point, such as when the original pattern is rotated, make subsequent recognition difficult.

Furthermore, finding a discontinuity and adding or subtracting 8 to the code complicates such processing.

〔the purpose〕

In view of the above points, it is an object of the present invention to eliminate the above-mentioned drawbacks, capture the outline of a pattern as continuous data, and provide an image processing device that is capable of stable pattern recognition unaffected by pattern displacement, rotation, etc. be.

〔Example〕

Hereinafter, the present invention will be described in detail with reference to the drawings. 5 Images in the present application are those showing characters, marks, symbols, pictures, etc. 0 Figure 5 is a diagram showing an example of an image processing device to which the present invention can be applied 0 It should be noted that such a structure There is no need to limit the company.

10 is a hand-held image processing device, light sources from 11 companies,
12 is a document to be read. In addition, with optical systems from 13 companies,
A row 5 processing section 15 guides light from the light source 11 and the pattern on the original 12 to a photoelectric conversion element (for example, a planar CCD) 14 and processes output data and/or outputs data from the photoelectric conversion element 14 .

Next, the processing section 15 will be explained in detail.
FIG. 1 is a block diagram of the processing section 15.

The zero-order CPU 17 and memory 18 extract the features of the pattern, which performs binarization, noise removal, etc. on the input pattern in the 16th pre-processing circuit. Regarding feature extraction, the flow is shown in Figure 6-2. Then, the contour information of the pattern is encoded by primary feature extraction. Depending on the pattern or branch extraction method, the code is used as is.

Further, the code string and the feature by function may be extracted by secondary feature extraction. Then, the features obtained in the above manner are checked against a dictionary in the memory 18 or the like, or the degree of similarity or correlation is calculated, and the pattern is determined. 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.

FIG. 7 shows the direction code at each contour point to which the present invention is applied. In the case of this embodiment, the code represents the direction at each contour point when tracing the contour of the pattern clockwise. Note that it is not necessary to limit the tracking direction to this.

Here, the direction code at a certain contour point is determined by setting the contour line direction at the previous contour point to 0. That is, when tracing the contour of a pattern clockwise, the direction code at a certain contour point Pi, the contour direction of Pi is
0 if it matches the contour line direction at the previous contour point Pi-1, and 45 to the right compared to the contour line direction of Pi-1.
．． If the rotation is 90, 135° and 180°, each will be 1.2
, 5.4, and vice versa, if the rotation is 45 degrees, 90 degrees, and 135 degrees to the left, they are determined as -1, -2, and -3, respectively.0 For example, in the case shown in Figure 8, the direction codes are ro, 0, 1, 0. ．． −2.0,1,
2. OJ is determined as 0. Here, the direction code for SPl cannot be determined unless the contour line direction of the previous contour point is known, but in general, when tracing the contour of a pattern, it always returns to the starting point, so The direction code of the starting point between those points is also determined.

By this method, when the direction code of the pattern shown in Fig. 2 is inserted, it becomes as shown in Fig. 9, as is clear from Fig. 0.
According to this method, the resulting code train will be continuous. Furthermore, even after going around the contour, the codes for the same contour point have the same value.

In other words, it has periodicity whose period is the number of contour points.

From this K, the obtained code string can be treated as a periodic function5. In this case, even if the starting point of contour tracking changes, only the phase of the periodic function shifts. For example, the second
If 82 in the figure is taken as the starting point, the code string starts from 82 in FIG. 9, but if it is regarded as a periodic function, the function only shifts in the horizontal axis direction, and the shape of the function remains unchanged.

Due to the above properties, the code string obtained by the present invention is
FIG. 6 is superior to that shown in the fourth factor.Next, FIG. 10 shows a flowchart for inserting the code string described above. In step 1, a starting point for contour tracing is determined. Next, in step 2, the next first contour point is found. In step 6, the contour tangent direction is determined between the starting point and the first contour point.

Next, in step 4, select the m-th contour point (m-2, 3
, 4, . . . ), and in step 5, the contour tangent direction is determined with the m-1th contour point. In step 6, the difference in the contour tangent direction is encoded. Next, in step 7, it is determined whether the process has returned to the starting point. If you have not returned to the starting point, repeat the above steps.

Now, pattern identification is done by detecting, for example, the degree of curvature of the outline (straight, 1 right turn, left turn, sudden right turn, sudden left turn, ... meeting ... pot) from the code string obtained here. can be done. For example, the contour information corresponding to FIG. 11 can be obtained as shown in FIG. represent.

Furthermore, since the code string can be treated as a periodic function, it is also possible to perform, for example, a discrete Fourier transform on the function and identify it on the frequency axis.

As described above, the method according to the present invention makes it possible to stably extract contour information of a pattern, facilitates the subsequent pattern identification process, and conversely limits the pattern recognition process. not.

In the above explanation, each contour point where the direction code is placed is adjacent to each other, but this is
As shown in 8, adjacent contour points may be sampled, for example, every second contour point, or a method of assigning direction codes to contour points separated by a certain distance may be used. According to this method, more stable contour features can be extracted without being affected by fine irregularities in the pattern contour.

Further, in the above description, the direction code is quantized in eight directions, but it may be subdivided as shown in FIG. 14, for example. The numbers in parentheses shown in Fig. 16 were determined according to the direction code shown in Fig. 14, and can more faithfully display the curvature of the contour. Suitable for pattern recognition.

It goes without saying that the methods described above may be repeated or may be combined.

〔effect〕

As described in detail above, in 5, when the contour of a pattern is traced in the direction along which the present invention passes, the difference between the contour line direction at each contour point and the contour line direction at the previous contour point is determined. By encoding, the characteristics of the outline of the pattern can be stably expressed, and a continuous and periodic direction code string can be easily obtained, which reduces the processing required for the overall pattern recognition, and It has become possible to provide an m-image processing device that can improve the weaving rate (9) and has the above-mentioned effects.

[Brief explanation of drawings]

FIG. 1 is a diagram showing a conventional direction code, FIG. 2 is a diagram showing what the pattern is, FIG. 6 is a diagram showing the relationship between contour points and direction codes, and FIG. Figure 6-1 is a block diagram of the processing unit; Figure 6-2 is a diagram showing the flow of pattern feature extraction; Figure 7; FIG. 8 is a diagram showing an example of a pattern; FIG. 9 is a diagram showing the relationship between contour points and direction codes; FIG.
The figures are a flowchart for assembling a code string, Figure 11 is a diagram showing the L1 pattern and outline information, Figure 12 is a diagram showing the code information and outline information of the pattern, and Figure 16 is a diagram showing an example of the pattern. , FIG. 14 is a diagram showing subdivided direction codes. 17 is a CPU, and 15 is a processing unit. Applicant: Canon Co., Ltd. Kabuto 1j Kasumigata 12

Claims (1)

[Scope of Claims] Detection means for tracing contour information of an image pattern in a predetermined direction and detecting contour points, and calculation means for determining contour tangential directions based on the plurality of contour points detected by the detection means. An image processing apparatus characterized in that the image processing apparatus further comprises means for sequentially determining and encoding the directional differences in the contour tangential directions.