JPH11232463A - Picture recognizing device and method therefor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To precisely and efficiently recognize a prescribed graphic in an input picture based on the feature quantity of the graphic by extracting them. SOLUTION: This picture recognizing device is provided with a picture vector detecting means 1 for detecting picture vectors which indicate the densities and/or the chromaticity change and the direction of picture edges in the respective pixel positions of the input picture as the feature quantity of the graphic in the input picture, a reference data vector generating means 4 for generating the picture vectors in the respective picture positions of a convex line graphic being an object to be recognized, a vector comparing means 2 for comparing the picture vectors of the respective pixels in the specified area of the input picture, which are obtained by the picture vector detecting means 1, with the picture vector in the position corresponding to reference picture vector data generated by the reference data vector generating means 4, and a judging means 3 for judging the presence or absence of the graphic being the object in the input picture, based on the comparison result of the vector comparing means 2.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to an image recognition apparatus and method for recognizing whether or not an input image includes a convex curve figure having a predetermined shape.

[0002]

2. Description of the Related Art Various types of image processing for extracting an image having specific characteristics from an image input from a scanner using a CCD (Charge Coupled Device) image sensor or an image generated by a personal computer or the like are known. . These extraction techniques are widely used as a pre-process for pattern recognition by generating various new image processes by performing various types of image processing on the extracted images.

[0003] JP-A-6-178098, JP-A-6-18098
In the technique disclosed in Japanese Patent Application Laid-Open No. 225134, an input image is binarized at a predetermined density or chromaticity, and a pitch corresponding to the diameter of a target circular image in a scanning line direction or a direction perpendicular thereto is detected. .

In the technique disclosed in Japanese Patent Application Laid-Open No. 6-245064, pattern matching is performed by reducing the resolution of an input image and blurring the image.

[0005] As a general image recognition method,
There is a technique disclosed in Japanese Patent Application Laid-Open No. Hei 7-160885. Here, the input image is subdivided into small area meshes, a chromaticity histogram is created for each mesh, and the chromaticity histogram (called a codebook) of the target image is compared to determine whether or not the target image is a figure to be recognized. Is determined.

[0006] Separately, a chain coding technique for finding the features of an image by detecting the perimeter of the image is referred to as the "Image Analysis Handbook" (published by the University of Tokyo, Mikio Takagi, and supervised by Hirohisa Shimoda), pages 611-612. Is disclosed. The technique disclosed in Japanese Patent Application Laid-Open No. H5-114027 detects a target image by using this chain code as a feature amount.

[0007]

However, the techniques disclosed in JP-A-6-178098 and JP-A-6-225134 pay particular attention to circles, and it is difficult to recognize other figures. In particular, it is difficult to distinguish the figure from a figure such as an ellipse or an egg.

In the technique disclosed in Japanese Patent Application Laid-Open No. 6-245064, since the image is blurred, the feature amount of the figure is significantly reduced, and the recognition accuracy is reduced.

In the technique disclosed in Japanese Patent Application Laid-Open No. Hei 7-160885, since a chromaticity histogram is used as a feature value, geometric information of a figure may be lost.

In the technique disclosed in Japanese Patent Application Laid-Open No. H5-114027, since only the perimeter is used as the feature value, if the perimeter is the same, a figure having a different shape may be recognized as the same. .

[0011] In any of these techniques, the feature quantity of a figure to be extracted is limited, and it is difficult to identify a similar figure.

The present invention has been made in consideration of the above problems, and has been made in consideration of the above problems. An image recognition apparatus capable of extracting a characteristic amount of a graphic and accurately and efficiently recognizing a predetermined graphic in an input image based on the extracted characteristic amount is provided. It is an object to provide a method.

[0013]

In order to solve the above-mentioned problems, an image recognition apparatus according to the present invention is an image recognition apparatus for recognizing whether a recognition target convex curve figure having a predetermined shape exists in an input image. Means for detecting an image vector indicating the density and / or chromaticity change and direction of an image edge at each pixel position of the input image; and each of specific regions in the input image obtained by the image vector detection means. Vector comparison means for comparing the image vector of the pixel with the image vector at the corresponding position of the recognition target convex curve figure, and the presence or absence of the recognition target convex curve figure in the input image is determined based on the comparison result of the vector comparison means. And a judging means.

According to this, the image vector detecting means detects an image vector indicating the density and / or chromaticity change and direction of the image edge at each pixel position. This image vector represents the direction of the curve or straight line of the figure in the image. That is, information unique to the geometric shape of the figure in the image is extracted. In the convex curve figure, the image vector changes slowly due to pixels, and changes with a constant tendency, which is preferable as a feature value to be extracted. By comparing the image vector of the specific area in the input image with the image vector of the convex curve to be recognized by the image vector comparing means, the similarity between the graphic of the specific area in the input image and the convex curve to be recognized can be determined. . The determination means outputs whether or not the recognition target convex curve figure is included in the input image based on the similarity.

[0015] The image processing apparatus may further include storage means for storing the image vector at each pixel position of the convex curve figure to be recognized which has been obtained in advance.

[0016] According to this, the image vector comparing means includes:
The image vector value of the convex curve figure to be recognized is read at high speed from the storage means, and is compared with the image vector of the input image.

The specific area to be compared is preferably a rectangular or circular area containing the convex curve figure to be recognized. In this case, a partial area in the input image is cut out as a rectangular or circular area and compared with the image vector of the recognition target graphic. That is, the comparison target area can be made smaller than the input image.

The vector comparing means may compare only the image vector of the pixel of the outer peripheral part among the pixels of the convex curve figure to be recognized. In this case, only the outer contour information among the graphic information of the convex curve graphic is extracted and handled.

If the judging means judges that there is a convex curve figure to be recognized, it may output the type and position of the convex curve figure. This output is used as an input to more detailed pattern recognition and image processing.

The image vector detecting means compares the amount of change in density and / or chromaticity change around each pixel with one or more thresholds to determine whether or not there is a change in density and / or chromaticity at each pixel position and the direction. May be determined.

By comparing the amount of change in density and / or chromaticity change around each pixel with a threshold value, the image vector is converted into a digital numerical value having a value within a predetermined range which can be easily handled by the subsequent vector comparing means. Is done.

Further, another image recognition apparatus of the present invention comprises: an image vector detecting means for detecting an image vector indicating a density and / or a chromaticity change and a direction of an image edge at each pixel position of an input image; Are scanned, and each pixel is compared with each image vector of the scanned pixel adjacent thereto, and it is determined whether or not the amount of change satisfies a predetermined continuous condition as a convex curve. The curve detecting means for detecting the position of the partial curve of the convex curve in the input image, and the end points of the partial curve detected by the curve detecting means are respectively traced, and both end points coincide, the partial curve is closed and the convex curve is closed. Closed point detecting means for detecting a position of a closed point which is a matching position of end points when forming a figure, length calculating means for calculating a curve length of the convex curve, and a curve detected by the length calculating unit Length and closure Determination means for determining whether the recognition target convex curve shape present in the input image based on the detected position of the closed point detection means may be one provided with.

As described above, the image vector indicates the direction of the line of the figure in the input image. Therefore, in a continuous curve, particularly a convex curve, the change due to the position of the image vector needs to satisfy a predetermined condition. By examining whether the image vector satisfies the predetermined condition between adjacent pixels, the position of the partial curve of the convex curve can be clarified. The position of the partial curve is checked while scanning the input image, and the end point position is tracked. Finally, the end points coincide with each other, and the closed point position, which is the coincidence position of the end points when this partial curve is closed and a convex curve figure is finally formed, is confirmed. This makes it possible to recognize a convex curve figure. By examining the correspondence between the graphic position of the convex curve figure and the length of the convex curve, it is confirmed whether the figure is a convex figure to be recognized.

On the other hand, an image recognition method according to the present invention corresponds to each of the above-described image recognition apparatuses, and is an image for recognizing whether or not a recognition target convex curve figure having a predetermined shape exists in an input image. A recognition method comprising: an image vector detecting step of detecting an image vector indicating a density and / or a chromaticity change and a direction of an image edge at each pixel position of an input image; A vector comparison step of comparing the image vector of each pixel in the specific region with an image vector at a corresponding position of the recognition target convex curve figure; and a recognition target convex curve figure in the input image based on the comparison result of the vector comparison step. And a judging step of judging presence or absence.

Further, the vector comparison step may compare the image vector of each pixel in the input image with the image vector at each pixel position of the convex curve figure to be recognized which has been obtained and accumulated in advance.

The specific area is preferably a rectangular or circular area containing the convex curve figure to be recognized.

In the vector comparison step, it is preferable to compare only the image vector of the pixel of the outer peripheral portion among the pixels of the convex curve figure to be recognized.

In the determining step, when it is determined that there is a convex curve figure to be recognized, the type and position of the convex curve figure may be output.

In the image vector detecting step, the presence and absence of the density and / or chromaticity change at each pixel position are compared by comparing the amount of change in the density and / or chromaticity change around each pixel with one or more threshold values. Is preferably determined.

Alternatively, the image recognition method according to the present invention comprises an image vector detecting step of detecting an image vector indicating a density and / or a chromaticity change and a direction of an image edge at each pixel position of an input image; By scanning each pixel, and comparing each image vector of each pixel with a scanned pixel adjacent thereto by determining whether or not the amount of change satisfies a predetermined continuous condition as a convex curve. The curve detection step of detecting the position of the partial curve of the convex curve included in the input image and the end points of the partial curve detected in the curve detection step are respectively tracked, and these end points match, and the partial curve is closed and convex. A closed point detection step of detecting a position of a closed point that is a matching position of end points when forming a curved figure, a length calculation step of calculating a curve length of a convex curve, and a length detection step. The line length and closed point detection step determination step of determining whether the recognition target convex curve shape in the input image based on the detected position of the closed point is present in, or those comprising a.

[0031]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Preferred embodiments of the present invention will be described below in detail with reference to the drawings. In addition, in order to facilitate understanding, the same reference numerals are given to the same components as much as possible in each drawing, and duplicate description will be omitted.

FIG. 1 is a block diagram of a first preferred embodiment of the present invention. Image vector detection means 1 for detecting an image vector indicating the density and / or chromaticity change and direction of an image edge at each pixel in an input image and outputting the image vector data; and a predetermined convex curve to be recognized Reference data vector generation means 4 for generating reference image vector data which is an image vector value at each pixel position of the figure, and vector comparison means 2 for comparing the image vector data of the input image with the reference image vector data at the corresponding position. And a judging means 3 for judging whether the figure in the input image matches the recognition target image based on the comparison result and outputting the judgment result.

Here, the convex curve is a closed curve, which is a plane curve whose intersection with a straight line (excluding a tangent line) intersecting the closed curve is always only two points, and a typical curve is a circle, an ellipse, an oval or the like. It is.

FIG. 2 shows a detailed block diagram of the inside of the image vector detecting means 1. A line buffer unit 11 consisting of the line buffer 11 ₁ to 11 _n for storing image information of a given scan line in an input image, a predetermined region in the input image by the stored information of the line buffer unit 11 based on (hereinafter, A window processing unit 12 for extracting image information of a window area) and an image vector calculation circuit 13 for calculating and outputting an image vector at a central pixel of the window area extracted by the window processing unit 12.

Here, the image vector used in the present invention will be described in detail with reference to FIGS.

FIG. 3A shows a circular image as an example of a recognition target, and FIGS. 3B and 3C show the circular images.
It is a partial enlarged view of each of (b). CC input image
When an image is read by a scanner such as D or an image is generated by a computer, the image is represented as an array of pixels arranged two-dimensionally. For this reason, when the outer peripheral portion of the circular image shown in FIG. 3A is enlarged, it is represented as a square pixel row as shown in FIG. 3C. FIG.
(H) has shown the enlarged view of the image in each position of a circumference. If the size of each pixel is sufficiently small relative to the radius of the circle, that is, if the resolution of the image is sufficiently fine, the change in chromaticity and density between adjacent pixels of the edge of the image is negligible. As a result, changes in the density and / or chromaticity of pixels orthogonal to the edge direction are also equal between pixels adjacent in the edge direction.

Here, in the 7 × 7 pixel column shown in FIG. 4A, for example, the density of the image of the pixel (1, A) is set to f
If expressed as (1, A), the density change in the direction orthogonal to the edge direction at the pixel (4, D) is f (4, G) −
f (4, A). The density change in this direction takes the maximum value among the density changes in each direction passing through the pixel (4, D), while the density change f (7, D)-in the edge direction which is a direction orthogonal to this. f (1, D) becomes almost zero.
Then, the density change f (2, G) -f (2, A) in the direction orthogonal to the edge in the neighboring pixels (2, D) and (6, D) in the edge direction of the target pixel (4, D) and f
(6, G) -f (6, A) is also f (4, G) -f (4,
A). This is shown in FIGS.
The same applies to other pixel columns shown in (h). By utilizing this property, the direction of the edge can be detected.

More specifically, taking the case of FIG. 4A as an example, first, the amount of density change in a direction assumed to be orthogonal to the edge passing through the pixel of interest (4, D) is compared with a predetermined threshold value TH1. I do.

[0039]

(Equation 1)

The direction orthogonal to the assumed edge is detected by using either or both of the maximum density change in the direction orthogonal to the edge and the zero density change in the edge direction. Can be.

Next, in the direction orthogonal to this direction, the density change amount in the same direction passing through the pixels (2, D) and (6, D) near the target pixel (4, D) is determined. And it is confirmed that the density is constant within a predetermined threshold value TH2.

[0042]

(Equation 2)

The direction assumed when this condition is satisfied is the direction orthogonal to the edge. Then, a value representing a direction orthogonal to the detected edge is output as an image vector value of the pixel. FIG. 5 shows an example thereof, and FIG.
As shown in (i), numbers 1 to 8 are assigned corresponding to the directions of density change perpendicular to the image edge of the pixel, here eight directions, and 0 is assigned when no image edge is detected. . By this method, FIG.
5 (a) to 5 (h) show the results of obtaining image vectors for eight peripheral areas of the circles shown in FIGS.
Here, the number of peripheral pixels used when obtaining the image vector of the target pixel is 7 × 7 pixels, and the image vector is represented in eight directions. However, in order to further improve the accuracy, the number of peripheral pixels may be increased or the image may be increased. 16 vector directions, 3
It may be further divided into two directions.

Next, the operation of this embodiment will be described.
FIG. 6 is a flowchart of the operation of the present embodiment, and FIG. 7 is a detailed flowchart of the image vector detecting operation. Here, an example will be described in which 7 × 7 pixels centering on the target pixel are cut out as peripheral pixels, that is, as window regions, and image vectors are detected in eight directions.

When an image including a figure to be recognized is input and image recognition processing is started, data of each pixel of the image is sent to the image vector detecting means 1 shown in FIG. The following is the calculation step (S1) of the image vector shown in FIG. 6, the details of which are shown in FIG.

Image information of a predetermined number n of lines of the input image data is stored in a line buffer 1 shown in FIG.
It is stored in the 1 ₁ to 11 _n. And the window processing unit 1
2, the image data of a window area of a predetermined number of pixels and a 7 × 7 pixel area around the target pixel are cut out (S
11). The image data of this window area is sent to the image vector calculation circuit 13, and each direction of the window area passing through the pixel of interest, that is, each direction shown in FIG. Excluding 4
A density change amount is calculated for the direction (S12). Next, it is determined whether or not the density change amount in the direction indicating the maximum density change amount exceeds the threshold value TH1 as shown in Expression (1) (S13). If the density or color of the recognition target graphic is already known, it may be confirmed that the density change amount is within a certain range with respect to the predetermined value, or the density value itself of the target pixel may be confirmed.

If the density change amount exceeds the threshold value TH1, the direction orthogonal to the direction indicating the maximum density change amount is set as the edge direction. A density change amount of two pixels in a direction orthogonal to the edge direction is calculated (S14). Then, as shown in Expression (2), it is determined whether the difference between the density change amount and the density change amount of the target pixel is smaller than a predetermined threshold TH2 (S15). If it is less than the threshold value, an image vector value corresponding to a direction orthogonal to the edge direction as shown in FIG. 5I is output (S16). On the other hand, the density change amount in the direction showing the maximum density change amount in S13 is the threshold value TH.
If the difference is equal to or smaller than 1 and if the density change of the neighboring pixel and the density change of the target pixel are equal to or larger than the threshold TH2 in S15, 0 is output as a vector value (S17). The above is the details of the image vector detecting step of S1 shown in FIG.

Returning again to the flowchart of FIG. 6, the output input image vector data is sent to the vector comparison means 2 shown in FIG. The vector comparison unit 2 reads out image vector data of the recognition target graphic from the reference data vector generation unit 4. FIG. 8 shows an example of image vector data of the recognition target graphic. FIGS. 8A, 8C, and 8E show examples of the recognition target graphic, and show circles having different outer diameters X, Y, and Z mm, respectively. FIGS. 4B, 4D, and 4F show the distribution of image vector values at each pixel position corresponding to the outer periphery of each of these circles. On the other hand, FIG.
Is an example of an input image. Here, a case is shown in which a circle having the same outer diameter X mm as the recognition target graphic in FIG. 8A is used as the input image. FIG. 2B shows the calculation result of the image vector. The vector comparing means 2 determines whether the reference vector data (see FIGS. 8B, 8D, and 9F) matches the image vector data of the input image (see FIG. 9B) between the pixels. The number to be performed is detected (S2). This may be performed on the entire area, or may be performed only on the outer peripheral portion of the recognition target graphic indicated by the reference data in order to simplify and speed up the processing. In this example, when processing is performed only on the outer peripheral portion indicated by the reference data, FIG.
And 20 of the 20 image vector data in FIG. On the other hand, the image vector data of FIG. 9B hardly matches the image vector data of circles having different outer diameters shown in FIGS. 8D and 8F. Therefore, only a circle having the same outer diameter can be reliably identified. When the input image is larger than the recognition target image, it is preferable to cut out a partial region of the input image as a window region and compare the window region with the recognition target image. With this configuration, the number of processing circuits or the number of computations in the comparison unit can be reduced, and the comparison computation can be performed at high speed.

The number of matches and the identification data of the matched figure are sent to the judging means 3. The judging means 3 compares the number of matches with a predetermined threshold (S3). If the match is equal to or greater than the threshold, it is judged that the input image contains the recognition target graphic, and the identification data and the position of the graphic are output. (S4).
On the other hand, if the number of matches is less than the threshold, it is determined that the recognition target graphic is not included in the input image (S
5). If this threshold value is the number of perfect matches, that is, 20 in the case of the recognition target graphic in FIG. 8B, it becomes impossible to recognize when the input image contains noise or overlaps with another graphic. For example, it is preferable to reduce the number to about 15, for example.

In the above description, a circular image has been described as an example, but other convex figures such as an ellipse and an oval can be similarly recognized.

Next, a second embodiment of the present invention will be described. FIG. 10 is a block diagram of the second embodiment. The image vector detecting means 1 is the same as in the first embodiment, and a description thereof will be omitted. The detected image vector data is obtained by scanning an input image and detecting a curved portion of a convex curve in the scanned image, and a line length calculating unit 7 for detecting the length of the curved portion. Sent to The curve detecting means 5 traces the end point of the detected curve portion, and detects a closed point which is a matching position of the end point when the end points match and the convex curve portion is closed to form a convex curve figure. The point detection means 6 is connected, and the closed point detection means 6 and the line length calculation means 7 are connected to the judgment means 8 for judging the presence or absence of the detection target image in the input image.

The operation principle of this device will be briefly described with reference to FIG. FIG. 11 is a simplified illustration of a figure being recognized. In the pixel array of the input image, each pixel in one row is read one by one in the main scanning direction, and when reading of one row is completed, reading of the next row in the sub-scanning direction is performed. Reading of pixel information is performed. The position of the first scanned pixel among the pixels constituting the recognition graphic is hereinafter referred to as a diagram start position. Then, the last position of the pixel constituting the recognition graphic among the pixels scanned so far is hereinafter referred to as a diagram end position. This diagram end position corresponds to the end points at both ends of the partial curve of the recognized convex curve. In the case of a convex curve figure, there are always two diagram end positions except for the diagram start position and a closed point described later. This is clear from the definition of the convex curve described above. Then, the position of the last scanned pixel among the pixels constituting the recognition graphic is hereinafter referred to as a closed point. Therefore, when scanning is completed up to the closing point,
It is also clear that the two diagram end positions coincide at this closed point.

The operation of this device will be described with reference to FIGS.

FIG. 12 is a flowchart of the operation of this apparatus. First, the diagram start position and the diagram end position are set to initial values, for example, positions outside the scan end position of the input image, and scanning in the input image is started (S20). Next, the image vector of the pixel to be scanned is calculated by the image vector detecting means 1 shown in FIG. 10 (S21). This step is the same as steps S11 to S17 shown in FIG. 7, and a description thereof will be omitted. Subsequently, it is determined by the curve detecting means 5 whether the target pixel is adjacent to the diagram end position (S22). This diagram end position exists in the screen of the input image when the current scanning position is after the diagram start position shown in FIG. Here, if it is at the adjacent position, it is determined whether or not the image vector of the target pixel is in a continuous relationship with the image vector at the diagram end position (S2).
3).

Here, FIG. 13 is a diagram showing the continuity condition of the image vector of each pixel near the diagram end position when the image vector is represented in eight directions. For example, in the upper end portion of the convex curve figure, the image vector of the pixel at the diagram end position is 7. The image vector of the pixel to the left of this pixel is 7
Alternatively, when the image vector of the right pixel is 6 or 7, the continuous condition as a convex curve is satisfied. Each table is a value that can be taken when adjacent pixels satisfy the continuous condition as a convex curve when the center of the table is the value of the image vector at the diagram end position. Here, the image vector of the pixel inside the diagram may be a value other than 0. This continuity relationship may be confirmed by, for example, confirming the continuity relationship of the pixel of interest using a predetermined area around the pixel of interest, for example, a 3 × 3 pixel window area.

If the target pixel satisfies the continuity relation with the diagram end position, the diagram end position on the continuous side is updated with the target pixel position (S24). Then, the line length from the diagram start position is calculated and updated by the line length calculating means 7 based on the target pixel position (S25).

On the other hand, if the target pixel does not satisfy the continuous relationship with the diagram end position, it is first checked whether or not the scanning of the adjacent pixel at the diagram end position has been completed (S2).
6). If the scanning of all the adjacent pixels has been completed, it is checked whether the diagram end position is not an open point (S27). The open point is an end point of the figure as shown in FIG. 14, and the presence of the open point indicates that the recognized curve is not a part of the closed curve but a part of the open curve. If the diagram end position corresponds to the open point, since the figure being recognized is not a closed curve, the diagram start position, the end position, and the length of the curve are cleared to their initial values (S28).

If it is determined in S22 that the target pixel is not adjacent to the diagram end position, or if the diagram end position is an open point, it is checked whether the current scanning position is appropriate as the diagram start position. (S29). Specifically, when the image vector is represented in eight directions, the image vector at the diagram start position is one of 6 and 7. If it is appropriate as the diagram start position, this pixel position is set as a new diagram start position and new diagram end position (S30). Then, regardless of whether or not it is appropriate as the diagram start position, the process proceeds to S32 described later.

After updating the line length in S25, the closed point detecting means 6 checks whether or not the target pixel is a closed point (S31). This can be easily checked by checking whether the two diagram end positions match at the target pixel position. When the continuity relation is confirmed using the window area as described above, the check of the closing point can be performed simultaneously with the confirmation of the continuity relation. If the target pixel is not a closed point, that is, if the scan to the closed point has not been completed, the process proceeds to S32, where the scan target pixel is updated and the next pixel is scanned, and the processing from S21 onward is repeated.

If the target pixel is a closed point, the judging means 8 checks whether the relationship between the diagram start position and diagram end position and the obtained line length matches the image to be recognized (S).
33). If these pieces of image information match, it is determined that they match the recognition target image, and the position is output (S3).
5) If they do not match, it is determined that they do not match the recognition target image (S34). After the determination, it is determined whether the scanning of the entire screen has been completed (S36). If the scanning has not been completed, the process proceeds to S32, and the next pixel is scanned. If the scanning has been completed, the entire process is completed.

According to the present embodiment, recognition processing can be performed simultaneously with pixel scanning when reading an image, so that processing can be performed at high speed and a processing circuit for image recognition and a processing circuit required for processing are required. The number of memories and the number of calculation processes can be reduced, and realization is possible with an inexpensive device. On the other hand, since image recognition is performed using the features of the figure, reliable image recognition is possible. Here, an example in which the image vector is expressed in eight directions has been described. However, by dividing this direction into smaller ones and strictly determining the continuity condition in accordance with the convex curve figure to be recognized, more accurate image recognition can be performed. .

Further, since the memory is reduced, even if the resolution is reduced in any one of the processing stages, the recognition accuracy is not extremely reduced, and good image recognition can be performed.

According to the present invention, for example, in order to prevent copying of paper money or securities such as stock certificates and receivables in a color copying machine, the position of the red seal in these securities is specified, and the red seal is recognized. In this case, the present invention can be applied to an image recognition device or the like for operating a function for preventing unauthorized copying.

Further, the present invention can be applied to preprocessing for recognizing the position of a specific convex curve figure for performing a specific image processing on the convex curve figure.

[0065]

As described above, according to the present invention,
By detecting the image vector of each pixel and comparing with the image vector of the convex curve figure to be recognized, it is detected whether or not the convex curve figure to be recognized is included in the input image. Therefore, since image recognition is performed using the characteristic amount specific to the geometric information of the convex curve figure, highly accurate and efficient detection can be performed.

Further, it is preferable to accumulate the image vectors of the convex curve figures to be recognized in advance because the comparison processing can be performed at high speed.

If a specific rectangular or circular window area is cut out from the input image and compared with the recognition target graphic, the number of processing circuits used for the comparison processing and / or the arithmetic processing at the time of the comparison processing can be reduced. Processing can be realized at low cost.

Since the processing can be simplified and the circuit and the like can be simplified by making only the image vector on the outer periphery of the figure as the object of the comparison processing, it is particularly preferable to use it as a preprocessing stage of pattern recognition.

It is preferable to output the type and position of the recognized convex curve figure because pattern recognition and image processing performed as post-processing are strengthened.

Further, when an image vector is determined, the amount of change in luminance and / or chromaticity change around the target pixel is determined by comparing it with one or more threshold values. Can be easily and reliably recognized.

On the other hand, according to another embodiment of the present invention, a convex curve figure is recognized by tracing a partial curve of the convex curve in the input image based on the detected image vector data. Since the amount of information specific to the geometric information is used, accurate image recognition can be performed. Further, since the processing can be realized by a simple processing circuit and / or arithmetic processing in accordance with the image reading, image recognition can be performed at high speed and at low cost.

[Brief description of the drawings]

FIG. 1 is an overall block diagram of a first embodiment of the present invention.

FIG. 2 is a block diagram of an image vector detecting unit of the embodiment of FIG.

FIG. 3 is a diagram showing image information of a figure to be recognized.

FIG. 4 is an explanatory diagram of image vector detection according to the present invention.

FIG. 5 is a diagram illustrating an image vector according to the present invention.

FIG. 6 is a flowchart of the operation of the embodiment of FIG. 1;

FIG. 7 is a detailed flowchart of an image vector detecting step in the operation shown in FIG. 6;

FIG. 8 is a diagram illustrating an example of a recognition target graphic and its image vector.

FIG. 9 is a diagram showing an example of an input image and its image vector.

FIG. 10 is an overall block diagram of a second embodiment of the present invention.

FIG. 11 is a diagram illustrating the operation principle of the second embodiment of the present invention.

FIG. 12 is a flowchart of an operation according to the second embodiment of the present invention.

FIG. 13 is a table showing a continuity relationship of a pixel as a convex curve.

FIG. 14 is a diagram showing an open point.

[Explanation of symbols]

1 ... image vector detecting means, 2 ... vector comparing means, 3
... determining means, 4 ... reference data vector generating means, 5 ... curve detecting means, 6 ... closed point detecting means, 7 ... line length calculating means,
8 ... determination means, 11 ... line buffer, 12 ... window processing unit, 13 ... image vector calculation circuit.

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Terui Hayashi 2274, Hongo, Fuji Xerox Co., Ltd., Ebina City, Kanagawa Prefecture ) Inventor Kenichi Takahashi 2274 Hongo, Ebina-shi, Kanagawa Fuji Xerox Co., Ltd. (72) Inventor Akihiko Fuya 2274 Hongo, Ebina-shi, Kanagawa Fuji Xerox Co., Ltd.

Claims (14)

[Claims] 1. An image recognition apparatus for recognizing whether or not a recognition target convex curve figure having a predetermined shape exists in an input image, comprising: a density and / or a chromaticity of an image edge at each pixel position of the input image. An image vector detecting means for detecting an image vector indicating a change and a direction, at a position corresponding to the image vector of each pixel of the specific region in the input image obtained by the image vector detecting means and the recognition target convex curve figure An image recognition apparatus comprising: vector comparison means for comparing an image vector; and determination means for determining whether or not the recognition target convex curve figure exists in an input image based on a comparison result of the vector comparison means. 2. The image recognition apparatus according to claim 1, further comprising storage means for storing an image vector at each pixel position of the convex curve figure to be recognized which is obtained in advance. 3. The image recognition apparatus according to claim 1, wherein the specific area is a rectangular or circular area including the convex curve figure to be recognized. 4. The image recognition apparatus according to claim 1, wherein the vector comparison unit compares only the image vector of an outer peripheral pixel among the pixels of the convex curve figure to be recognized. 5. The image recognition apparatus according to claim 1, wherein said judging means outputs the type and position of the convex curve graphic when the judgment is made that the convex curve graphic to be recognized exists. 6. The image vector detecting means compares the amount of change in density and / or chromaticity change around each pixel with one or a plurality of thresholds to determine the change in density and / or chromaticity change at each pixel position. 2. The image recognition apparatus according to claim 1, wherein presence / absence and direction are determined. 7. An image recognition apparatus for recognizing whether or not a recognition target convex curve figure having a predetermined shape exists in an input image, comprising: an image edge density and / or chromaticity change at each pixel position of the input image. An image vector detecting means for detecting an image vector indicating an image vector and a direction, and scanning pixels constituting an input image, and comparing each image vector of each pixel with a scanned pixel adjacent thereto, and calculating a change amount thereof. Curve detecting means for detecting the position of a partial curve of the convex curve included in the input image by determining whether or not a predetermined continuous condition as a convex curve is satisfied, and a portion detected by the curve detecting means Closed point detecting means for tracking each end point of the curve, the end points being coincident, and detecting a position of a closed point which is a coincidence position of the end points when the partial curve is closed to form a convex curve figure, Length calculating means for calculating the curve length of the convex curve, and in the input image based on the curve length detected by the length calculating unit and the closed point position detected by the closed point detecting means Determining means for determining whether or not the convex curve figure to be recognized exists; 8. An image recognition method for recognizing whether or not a recognition target convex curve figure having a predetermined shape exists in an input image, comprising: a change in image edge density and / or chromaticity at each pixel position of the input image. And an image at a position corresponding to the image vector of each pixel of a specific region in the input image obtained by the image vector detection step and the recognition target convex curve figure. An image recognition method comprising: a vector comparison step of comparing a vector with a vector; and a determination step of determining whether or not the recognition target convex curve figure exists in an input image based on a comparison result of the vector comparison step. 9. The vector comparison step of comparing an image vector at each pixel position of the convex curve figure to be recognized and obtained in advance with an image vector of each pixel in the input image. Image recognition method. 10. The image recognition method according to claim 8, wherein the specific area is a rectangular or circular area including the convex curve figure to be recognized. 11. The image recognition method according to claim 8, wherein in the vector comparison step, comparison is performed only with an image vector of a pixel in an outer peripheral portion among pixels of the convex curve figure to be recognized. 12. The image recognition method according to claim 8, wherein in the determining step, when it is determined that the convex curve graphic to be recognized exists, the type and position of the convex curve graphic are output. 13. The image vector detecting step according to claim 1, wherein a change in density and / or chromaticity change around each pixel is compared with one or a plurality of threshold values to determine a change in density and / or chromaticity change at each pixel position. 9. The image recognition method according to claim 8, wherein presence / absence and direction are determined. 14. An image recognition method for recognizing whether or not a convex curved figure having a predetermined shape exists in an input image, comprising: a density and / or chromaticity change and a direction of an image edge at each pixel position of the input image. An image vector detecting step of detecting an image vector indicating the following.Scanning pixels constituting the input image, comparing each pixel with an image vector of a scanned pixel adjacent thereto, and calculating a change amount of the pixel. A curve detecting step of detecting a position of a partial curve of a convex curve included in the input image by determining whether or not a predetermined continuous condition as a curve is satisfied; and A closed point detection step of tracking each end point, detecting the position of a closed point that is a coincidence position of the end points when the end points coincide and the partial curve is closed to form a convex curve graphic; and A length calculating step of calculating a curve length of the line; and the recognition in the input image based on the curve length detected in the length calculating step and the position of the closed point detected in the closed point detecting step. A determining step of determining whether or not the target convex curve figure exists;