JPH0410113B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field of the Invention] The present invention relates to a figure recognition device that can recognize a figure by stably extracting contour information of the figure.

[Technical background of the invention and its problems]

It is known that one method of recognizing a figure is to use the contour information of the figure. The contour information of the figure is given as a direction code string obtained by tracing the contour of the figure and quantizing the contour tracing direction (tangential direction of the contour line) at each contour point into, for example, four directions or eight directions. This method is called a Freeman code, and for example, according to a predetermined direction code as shown in FIG. 1, the outline of the figure pattern shown in FIG. It is something that will be coded. Therefore, in this example, the direction code string of the outline of the graphic pattern is (1, 2, 1, 2, 2, 2...2, 3, 6,...5,
6, 5, 6, 0). Such a direction code string preserves information regarding the original figure, and therefore has the property that the original figure can be completely reproduced from this direction code string.

However, since the quantization accuracy in the contour tracing direction to obtain the direction code is generally low, it is sensitive to irregularities occurring at one contour point, and has the disadvantage of being easily affected by noise such as so-called blurring of figure edges. ing. Furthermore, due to this problem, there was a problem in that it was difficult to use the direction code string as it was as feature information of the figure outline in the recognition process.

Incidentally, smoothing processing is generally known as a means for reducing the influence of noise. However, in the direction code string indicating the outline of a figure, the direction code changes from "0" to "7", or vice versa, from "7" to "1".
Conventionally, the above-mentioned smoothing process for Freeman codes is inappropriate because there are discontinuous points that change to This has been considered to be an obstacle to the development of figure recognition processing using contour information.

[Purpose of the invention]

In view of these circumstances, the present invention has been made with the focus on the fact that it is not necessarily necessary to reproduce the original figure from the standpoint of figure recognition, and its purpose is to reproduce the figure outline. To provide a highly practical figure recognition device capable of eliminating discontinuity in a direction code string to enable effective smoothing processing thereof and stably extracting useful information on the outline of a figure.

[Summary of the invention]

The present invention uses an analysis base that determines N different directions and assigns a direction code value to each direction with a certain difference between adjacent directions. In a figure recognition device that sequentially observes direction code values in the contour line direction between contour points based on the above analysis base, and recognizes the figure to be recognized from the obtained direction code value sequence, the observed figure at the current observation point is used. A processing rule is created based on both input values and the analysis base by inputting the direction code value that has been obtained and the direction code value that has already been determined by being observed at the observation point one time before the current observation point. Therefore, the correction value obtained by adding the difference value to the maximum direction code value assigned to the N directions is added to the direction code value observed at the current observation point. , subtraction, or no addition or subtraction, executes one of the processes, and outputs the result as a determined direction code value at the current observation point.

〔Effect of the invention〕

Thus, according to the present invention, the direction code string quantized in eight directions is "6", "7", "0", "1", for example.
If it is given as ``6'' or ``7'',
It becomes possible to capture the numbers "8" and "9" continuously, and by stably and reliably smoothing them, it is possible to effectively extract the features of the figure contour and perform stable figure recognition. In other words, since the discontinuity of the direction code is removed, it is possible to easily obtain a direction code string that strongly reflects the characteristics of the figure contour through smoothing processing.

[Embodiments of the invention]

Hereinafter, one embodiment of the present invention will be described with reference to the drawings.

Now, it is assumed that the contour of the figure shown in FIG. 2 is traced clockwise according to the direction code quantized as shown in FIG. 1, and the direction code at each contour point is sequentially determined. The creation of this direction code at each contour point is
For example, by focusing on the direction in which pixels forming the contour of a figure are lined up and detecting a 3×3 pixel area centered on the pixel of the contour point of interest using a mask or the like, the contour tracing direction is detected; This is done by quantizing this.

However, the arrangement of the outline pixels forming the figure outline is limited to six ways, as shown in FIGS. 3a to 3f, depending on the nature of the figure. Note that the number assigned to the pixel (i
-1), i, and (i+1) indicate the order of contour tracing, and i is the pixel of the contour point currently being focused on. In this example, each case is shown in which direction the pixel (i+1) moves after the pixel is tracked from pixel (i-1) to pixel i in the direction indicated by the direction code "5".
In this case, when the direction from pixel i to pixel (i+1) is as shown in FIG. Furthermore, if there is no change in the direction code as shown in FIG. 3c, it can be determined that there is no curve in the figure outline. Furthermore, the third
Figures d to f show cases where the figure outline has a convex curve, and the direction code "5" is "6".
It changes from "7" to "0".

As is clear from such changes in the direction code, the contour shape of the figure is generally indicated depending on whether the value of the direction code increases, decreases, or does not change. However, as shown in Figure 1, there are only eight direction codes that can be quantized, so in the case of the convex figure outline shown in Figure 3, the direction code is from "5" to "5". Since the value of the direction code changes to "0" and the value of the direction code decreases, the figure contour is captured as if it had a concave curve. In other words, it can be understood that the change in direction code is discontinuous. There are only a limited number of shapes of figure contours that cause such discontinuity in the direction code, and there are 10 shapes shown in FIGS. 4a to 4j. Therefore, in such a case, the value of the direction code obtained by quantizing the contour tracing direction is not directly adopted, but the direction indicated by the direction code "0" is changed to the equivalent direction code "8". ”, it is possible to stably capture this as a figure contour having a convex curve that changes from direction code “5” to “8”.

The apparatus of the present invention performs the above-described processing on the direction code obtained at each contour point of a figure, thereby eliminating discontinuity in the direction code string. By this, for example, the direction code string shown in FIG. 5, which is obtained by tracing the outline of the figure shown in FIG. gender is removed. Therefore, by using a direction code sequence from which such discontinuities have been removed, it is possible to smooth the sequence and effectively extract the features of the figure contour.

That is, if the direction code sequence from which discontinuities have been removed is smoothed, the rate of change of the direction code will correspond to the curvature of the outline of the original figure. Therefore,
By focusing on the curvature of each contour point, it is possible to find straight parts, curved parts, etc. of a figure outline, and to effectively extract features of the figure outline such as concave parts, convex parts, and edges, and perform recognition processing. becomes.

Now, the above-mentioned process for removing discontinuity in the direction code string is performed, for example, as follows. FIG. 7 is a schematic diagram of the processing circuit. Reference numeral 1 designates a latch circuit into which a direction code obtained by quantizing the contour tracing direction at each contour point is input. The direction code latched in the latch circuit 1 is transferred to the subsequent latch circuit 2 and stored at the next timing. Therefore, the latch circuit 1 stores the direction code of the contour pixel i, and the latch circuit 2 stores the direction code of the preceding contour pixel (i-1).

The ROM 3 that generates the function F receives the direction codes stored in the two latch circuits 1 and 2 as address data, and inputs the direction codes stored in the two latch circuits 1 and 2 as address data.
A three-value function F consisting of 1, 0, and -1 is output. This function value F is applied to a latch circuit 5 via an adder 4 and is successively accumulated. The output data of the adder 4 is added to the direction code stored in the latch circuit 1 as its equivalent conversion control information and output. It should be noted that latch circuits 1, 2, and 5 receive the same clock signal CLK and respond to tracing the figure outline.
The latch is driven simultaneously at each contour point timing.

The ROM 3 stores a function F for determining the direction code at each contour point, and the direction code is determined in the following manner.
Now, the direction code found for contour pixel i is
C ^* _i , if the contour direction code already determined for the previous adjacent contour pixel (i-1) is C _i-1 , the contour direction of the contour pixel i obtained while preserving continuity. Code Ci is determined as follows.

Ci=C ^* _i [Ci _-1 /8]*8 +F(MOD(Ci _-1 , 8), C ^* _i However, in the above formula, MOP(a, b) is the remainder when a is divided by b, [a] is the integer part of a rounded down to the decimal point. F(a, b) indicates the number determined by the values of a and b. Here, the conditions shown in Figure 4 a to j described above are used. From b=0, 4≦a≦7... F(a, b)=8 1≦b≦2, 6≦a≦7... F(a, b)=8 b=7, 0≦a≦1... F (a, b) = -8 Others... The function F (a, b) = 0 is given. The function F +1, 0, -1 that controls this function F (a, b) is It is stored in the ROM 3 and read out according to the values of a and b, that is, the values of direction codes Ci and C _i-1 .

According to the device configured in this way,
In cases where discontinuity occurs in the direction code string,
Direction code C ^* _i found at its contour point i
8 or -8 is added to the contour direction code.
Ci will be determined. Further, when the figure has a spiral shape as shown in FIG. 8, the above-described addition process is further repeated depending on the number of occurrences of the discontinuous points. Therefore, the continuity of the direction code sequence is maintained in any case, and by smoothing this, it is possible to stably extract the contour features of the figure and perform its recognition with high precision. . Therefore, its practical advantages are extremely high;
A tremendous effect can be produced.

Note that the present invention is not limited only to the above embodiments. For example, the data to be added to the direction code
The device may be configured to read directly from the ROM 3. Further, even when tracing a figure contour counterclockwise, the same can be achieved, and the method of determining the direction code is not limited to the example shown in FIG. 1. In short, the present invention can be implemented with various modifications without departing from the gist thereof.

[Brief explanation of drawings]

FIG. 1 is a diagram showing an example of a direction code, FIG. 2 is a diagram showing the relationship between a figure pattern and its outline tracing, and FIGS. 3 to 8 are diagrams showing an embodiment of the present invention. Figures a to f are diagrams showing changes in contour pixels, Figures 4 a to j are diagrams showing contour pixel changes in which direction codes become discontinuous, Figure 5 is a diagram showing quantization direction code strings, and Figure 6. 7 is a diagram showing a contour line direction code string with discontinuities removed, FIG. 7 is a diagram showing the main part of the apparatus of the embodiment, and FIG. 8 is a diagram showing a spiral pattern figure and its direction code. 1, 2, 5...Latch circuit, 3...ROM (function F), 4...Adder.

Claims (1)

[Claims] 1. Tracking the outline of a figure to be recognized using an analysis base that determines N different directions and assigns a direction code value to each direction with a certain difference between adjacent directions. In a figure recognition device that sequentially observes the direction code values in the contour line direction between each contour point based on the above analysis base, and recognizes the figure to be recognized from the obtained direction code value sequence, the current observation The direction code value observed at the current observation point and the direction code value observed at the observation point one time before the current observation point and which has already been determined are input, and the direction code value is created based on both input values and the analysis base. A correction value obtained by adding the difference value to the maximum direction code value assigned to the N directions with respect to the direction code value observed at the current observation point according to the processing rule, It is characterized by comprising processing means that determines whether to add, subtract, or do not add or subtract, executes one of the processes, and outputs the result as a determined direction code value at the current observation point. Shape recognition device. 2. The processing means divides the direction code value observed and already determined at the observation point one time before the current observation point by the correction value, and calculates the value of the integer part of the value obtained by this division. Claim 1, further comprising a correction means for correcting the determined direction code value by adding a value obtained by multiplying the fixed direction code value by the correction value. The shape recognition device described.