JP3244041B2 - Image recognition method, image recognition device, and recording medium - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image recognition method, an image recognition apparatus, and a recording medium for recognizing a specific image such as a figure, character, or symbol from an input image.

[0002]

2. Description of the Related Art Conventionally, there is a template matching method as a method for recognizing a specific image from an input image. The procedure of a general template matching method is as follows. An image to be recognized is prepared as a reference image, and the input image and the reference image are superimposed and compared while shifting one pixel at a time in the horizontal or vertical direction. If all or a certain number or more of the pixels match, the recognition is performed. It is determined as the target image. However, in this method, the orientation of the image to be recognized is changed (rotated).
There was a disadvantage that it could not be recognized once it was entered. As a method for solving the above problem, as described in Japanese Patent Application Laid-Open No. Hei 5-12446, template matching is performed at various angles by generating coordinates when a reference pattern stored in a memory is rotated. A method has been proposed. Further, as described in JP-A-8-63604, a method of detecting the center of gravity of an image, converting the image into polar coordinate data centered on the center of gravity, and corresponding to a rotated image, and Japanese Patent Application Laid-Open No. 6-309461. As described in Japanese Patent Application Laid-Open No. H11-163, there has been proposed a method in which a Fourier transform is performed on a gray-scale pixel data string on a circular window to obtain a spectrum, which is compared with a spectrum pattern of a reference image and corresponds to a rotated image.

[0003]

However, in the prior art, there has been a problem that complicated calculations have to be performed as described above, and high-speed processing cannot be performed. In addition, although high speed can be expected if realized by a circuit, there is a problem that the circuit becomes large and complicated. On the other hand, in the conventional general template matching method, in addition to the above-described problem, in order to recognize a specific image, a window region having the number of pixels equal to or larger than the number of pixels of the specific image must be prepared. There is also a problem that the circuit becomes large-scale when the number of pixels of the image is large. By the way, in recent years, a specific pattern such as a two-dimensional code is preliminarily embedded in a work, important document, securities, or the like, and this pattern is recognized and used, so that advanced management and unauthorized use of the document (for example, unauthorized use of the work is prohibited). There has been proposed a system for realizing prevention of (copying) and the like, and a technology capable of recognizing a specific image faster than a general pattern matching method is expected. The present invention has been made in view of the above-described circumstances, and provides an image recognition method, an image recognition device, and a recording medium capable of quickly recognizing a specific image with a simple configuration and processing. Aim.

[0004]

In order to solve the above-mentioned problems, an image recognition method according to the present invention is a method for recognizing a specific image composed of a plurality of partial images having the same shape and the same size from an input image. An input image for recognizing the partial image from the input image and generating a binary image in which a pixel at a position where the partial image exists is an effective pixel, and a pixel at a position where the partial image does not exist is an invalid pixel A partial determination step, the binary image generated in the input image partial determination step is divided by a block having a size including a plurality of pixels, a block including the effective pixel is an effective block, and a block not including the effective pixel. An input image sampling step of simplifying the input image into a binary image expressed as an invalid block; and the input image sampling step of simplifying the input image sampling step. Matching the reference pattern representing a simplified advance in units of blocks the specific image as the image is characterized by having a total determination step of outputting a result of the matching. That is, since matching is performed by simplifying the image, the processing amount of the entire recognition process can be reduced. In addition, since the matching process is performed in block units after performing the recognition process in pixel units for each of the partial images, it is possible to suppress a decrease in recognition accuracy.

[0005] In addition, in the input image sampling step, the binary image generated in the input image portion determination step is divided into blocks each having a size including a plurality of pixels, and the block including the effective pixels is set to be valid. A block, a block that does not include the effective pixel is represented as an invalid block, and one valid block is represented by one valid pixel, and one invalid block is represented by one invalid pixel to generate a binary image. In the determining step, the binary image generated in the input image sampling step and the specific image are represented in advance in the block unit, and then each block is matched with a reference pattern in which each block is represented by one pixel. May be output. Thereby, the processing amount of the entire recognition processing can be significantly reduced.

It is to be noted that the partial image is recognized from the specific image, and a binary image is generated in which a pixel at a position where the partial image exists is an effective pixel and a pixel at a position where the partial image does not exist is an invalid pixel. The binary image generated in the image part determination step and the specific image part determination step is divided by the block, and a block including the effective pixel is expressed as an effective block, and a block not including the effective pixel is expressed as an invalid block. And a specific image sampling step of generating a reference pattern which is a binary image as described above, and the specific pattern may be obtained by these processes.

In the case where a plurality of effective pixels are adjacent to each other on the binary image generated in the input image part determining step (and the specific image part determining step), another effective pixel except one effective pixel is removed. An input image filtering step of changing the binary image as a pixel as an invalid pixel is provided, and the above-described processing is performed in the input image sampling step on the binary image changed in the input image filtering step. Is also good. As a result, the number of effective pixels for one partial image is “1”, so that it is possible to avoid a situation where there are a plurality of effective blocks corresponding to one partial image. Therefore, the recognition accuracy can be further improved.

Alternatively, when a plurality of effective pixels are adjacent to each other on the binary image simplified in the input image sampling step (and the specific image portion determining step), other effective pixels except one effective pixel are removed. An input image filtering step of changing the binary image with pixels as invalid pixels may be provided, and the above-described processing may be performed on the binary image changed in the input image filtering step in the overall determination step. Good. Also in this case, it is possible to avoid a situation where there are a plurality of effective blocks for one partial image. Therefore, improvement in recognition accuracy can be expected. Further, since the processing in the input image filtering step is performed in block units, there is an advantage that the processing amount in the processing can be reduced.

An input image binarizing step of binarizing the input image based on preset conditions to generate a binary image (and binarizing the specific image based on the color of the partial image) A specific image binarizing step of generating a binary image), and performing the processing of the input image part determining step (and the specific image part determining step) on the binary image generated in the binarizing step. If it is performed, the present image processing method can be applied to an image in which the density value of a pixel is multi-valued.

Further, in the above-described various methods, if the shape of the partial image is a shape in which the rotated image completely overlaps itself, even if the image to be processed is rotated and input, the difference is absorbed. be able to. Therefore, in the recognition processing of the partial image, it is not necessary to perform the recognition processing for each rotation angle. That is, it is sufficient to perform the partial image recognition process for any one rotation angle.

According to another aspect of the present invention, there is provided an image recognition apparatus for recognizing a specific image composed of a plurality of partial images having the same shape and the same size from an input image. A storage unit that stores a reference pattern set in advance according to the specific image, and recognizes the partial image from an image represented by input image data,
A partial determination unit that generates binary image data representing a binary image in which a pixel at a position where the partial image exists is an effective pixel, and a pixel at a position where the partial image does not exist is an invalid pixel; The binary image represented by the obtained binary image data is divided by a block having a size including a plurality of pixels, a block including the effective pixel is regarded as an effective block, and a block not including the effective pixel is regarded as an invalid block. Sampling means for simplifying the image into a represented binary image and generating binary image data representing the binary image, and reference pattern data representing a reference pattern representing the specific image simplified and represented in block units are stored. Storage means; and the second means generated by the sampling means.
An overall determination unit that matches a binary image represented by value image data with the reference pattern represented by the reference pattern data stored in the storage unit, and outputs a result of the matching. The image data input by the partial determination means is image data representing the input image. That is, since matching is performed by simplifying the pixels, it is possible to reduce the entire processing amount and the circuit scale. In addition, since the matching process is performed in block units after performing the recognition process in pixel units for each of the partial images, it is possible to suppress a decrease in recognition accuracy.

[0012] In addition, the sampling means may include:
A binary image represented by the binary image data generated by the partial determination unit is divided by a block having a size including a plurality of pixels, and a block including the effective pixel is an effective block and does not include the effective pixel. A block is represented as an invalid block, and one valid block is defined as 1
One effective pixel, the one invalid block is simplified to a binary image represented by one invalid pixel, and binary image data representing the binary image is generated. It is also possible to store reference pattern data representing a reference pattern in which each block is represented by one pixel after being simplified and expressed in units.
As a result, the overall processing amount and circuit scale can be significantly reduced.

When a plurality of effective pixels are adjacent to each other on the binary image represented by the binary image data generated by the partial determining means, another effective pixel excluding one effective pixel is removed. Filtering means for changing the binary image data as invalid pixels may be provided, and the sampling means may process the binary image data changed by the filtering means. As a result, the number of effective pixels for one partial image is “1”, so that it is possible to avoid a situation where there are a plurality of effective blocks corresponding to one partial image. Therefore, the recognition accuracy can be further improved.

Alternatively, when a plurality of valid pixels are adjacent to each other on the binary image represented by the binary image data generated by the sampling means, other valid pixels except one valid pixel are invalidated. Filtering means for changing the binary image data as pixels may be provided, and the binary judgment data changed by the filtering means may be processed by the overall judgment means. Also in this case, it is possible to avoid a situation where there are a plurality of effective blocks for one partial image. Therefore, improvement in recognition accuracy can be expected. Further, since the processing by the filtering means is performed in block units, there is an advantage that the overall processing amount and circuit scale can be reduced.

It is to be noted that the image data input by the partial judging means is the input image and the specific image, and the storage means stores the image data inputted by the partial judging means as image data representing the specific image. Alternatively, the binary image data simplified or generated by the sampling unit or the filtering unit may be stored as the reference pattern data. That is, the image recognition device may have a function of generating the reference pattern data from the specific image.

The input image data is binarized based on preset conditions to generate binary image data.
A binarizing unit, wherein the binarizing unit generates the binarizing unit;
The processing by the partial determination unit may be performed on the value image data. By doing so, the present image recognition apparatus can be applied to an image having a multi-valued pixel density value.

Alternatively, for the input image data,
A binarizing unit that performs binarization independently for each of the conditions based on a plurality of preset conditions to generate a plurality of binary image data; and the plurality of binarized images generated by the binarizing unit. Selecting means for generating one piece of binary image data based on the binary image data;
The processing of the partial determination means may be performed on the value image data. In this case, the image recognition device can be applied to an image in which the density values of pixels are multi-valued and the colors of the partial images are different.

Further, in the various apparatuses described above, if the shape of the partial image is a shape in which the rotated image completely overlaps itself, even if the image to be processed is rotated and input, the difference is absorbed. be able to. Therefore, in the recognition processing of the partial image, it is not necessary to perform the recognition processing for each rotation angle. That is, it is sufficient to perform the partial image recognition process for any one rotation angle, and the circuit scale can be reduced.

In order to solve the above-mentioned problem, a recording medium according to the present invention is a program for realizing an image recognition method for recognizing a specific image composed of a plurality of partial images having the same shape and the same size from an input image. An input image portion for recognizing the partial image from the input image and generating a binary image in which a pixel at a position where the partial image exists is an effective pixel and a pixel at a position where the partial image does not exist is an invalid pixel The determination step, the binary image generated in the input image portion determination step is divided by a block having a size including a plurality of pixels, a block including the effective pixel is a valid block, a block that does not include the effective pixel An input image sampling step for simplifying to a binary image expressed as an invalid block; The specific image and the binary image simplifying beforehand in the units of blocks matching a reference pattern expressed in, and characterized by recording a program for realizing the entire determination step of outputting a result of the matching. By executing the program recorded on the recording medium by a computer system or the like, the image is simplified and matching is performed, so that the processing amount of the entire recognition process can be reduced. In addition, since the matching process is performed in block units after performing the recognition process in pixel units for each of the partial images, it is possible to suppress a decrease in recognition accuracy.

[0020]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, an image recognition device (hereinafter, this device) according to an embodiment of the present invention will be described with reference to the drawings.

(1) Image Recognition Method First, an outline of an image recognition method (hereinafter, this method) realized by the present apparatus will be described. FIGS. 1 and 2 are flowcharts showing the flow of processing in the present method. In the present embodiment, the recognition processing and registration processing shown in these figures complete the recognition processing of a specific image with respect to an input image. Here, before describing a specific processing flow, the principle of the recognition processing in the present apparatus will be described.

(1-1) Basic Principle The present apparatus is based on the premise that a specific image is composed of a plurality of partial images, and each partial image has the same shape. That is, all partial images can be recognized by matching processing with one pattern, and can be recognized by examining the positional relationship of the partial images in an area of a size corresponding to the specific image based on the positions of the recognized partial images. Such a specific image is targeted for recognition.

In the present apparatus, first, a specific image to be recognized is subjected to partial image recognition processing to specify the position of the partial image in the specific image, and position data (hereinafter referred to as “position data”) roughly representing the position by a predetermined algorithm. , Reference image position data). The input image is subjected to the partial image recognition processing to specify the position of the partial image in the input image, and position data (hereinafter, input image position data) roughly representing the position by the predetermined algorithm.
Is determined, and the presence or absence of the specific image is determined by determining whether or not the image represented by the reference image position data exists on the image represented by the input image position data.

That is, the recognition process of the partial image is separated from the process of confirming the positional relationship of the partial images in the area of the size corresponding to the specific image. By using the image position data and the input image position data, the processing amount and the circuit scale related to the specific image recognition processing are reduced while ensuring the recognition accuracy. Further, in the present apparatus, the pattern used for the partial image recognition processing and the partial image of the specific image itself are formed into a pattern (for example, a circle or a square) having a shape completely overlapping with the own rotated image.
This makes it easy to handle rotated images.

(1-2) Image Recognition Method (1-2-1) Registration Process Next, the registration process in the present method based on the above-described principle will be described with reference to FIG. In the registration process in this method, first, in step SA1, a color data string (specific image data) of a specific image is input. After that, a plurality of binarization processes are performed in parallel on the specific image data input in step SA1, but here, in order to avoid a complicated description, step SA1 is performed.
The following description focuses on one binarization process starting from 2-1. In step SA2-1, it is determined whether each color data of the input image data is a specific color. The specific color here is the color of the partial image of the specific image,
In A2-1, a determination having a certain width is performed in consideration of an error. If the determination result is “YES”, the value of the pixel on the binary image data at the position corresponding to the pixel of the color data becomes “1” (step SA3-1a), and otherwise, the value of the same pixel Becomes '0' (step SA3
-1b). That is, binary image data in which the value of a pixel corresponding to a pixel of a specific color is “1” and the value of a pixel corresponding to a pixel that is not a specific color is “0” is obtained. In another binarization process performed in parallel (such as a binarization process starting from step SA2-2), the same process as described above is performed for another specific color. Note that a pixel with a value of “1” is a black pixel, and a pixel with a value of “0” is a white pixel.

Next, a selection process is performed in step SA4. Specifically, data corresponding to the specific image to be recognized is selected from a plurality of binary image data obtained as a result of each of the above-described processes, and is synthesized. For example, when the specific image to be recognized has a plurality of partial images and the colors of the respective partial images are different, a plurality of binary image data are selected,
Synthesized. This combination is performed, for example, by ORing a plurality of binary image data. As a result, binary image data corresponding to all the partial images exists on the binary image data.

In a plurality of steps starting from step SA5, a filtering process is performed. Specifically, step S
The binary image data obtained in A4 is scanned by, for example, a region R (see FIG. 3) covering the pixel of interest (X) and its surrounding pixels (A) to (D) to obtain the pixel of interest. The value of (X) is '1' and the other peripheral pixels (A) to
When at least one value of (D) is '1',
The value of the pixel of interest (X) is set to '0' (step SA7-
a), otherwise, the value of the pixel of interest (X) is not changed (step SA7-b).

FIG. 4 shows how the binary image data changes in the filtering process. In FIG. 4, an image represented by binary image data before filtering is shown in the upper part, an image represented by binary image data after filtering is shown in the lower part, and a region at a specific time point during the filtering is shown in the middle part. The images in R are shown, and in each image, the same pixels are denoted by the same reference numerals (P1 to P5). In this figure, scanning by the region R is performed from the upper left to the lower right in the figure with respect to the image before the filtering process. The main scanning direction is the horizontal direction in the figure, and the sub-scanning direction is the vertical direction in the figure.

FIG. 5 is a flowchart showing a specific flow of the filtering process using the region R. As shown in FIG. 5, in the filtering process, first, it is determined whether or not the target pixel is an effective pixel (black pixel) (step SB1).
For example, when the target pixel in the region R is P1 (see FIG. 4), the determination result in step SB1 is “NO”, and the next target pixel is selected without changing the value of the target pixel (step SB3-b). , SB4). When the target pixel in the region R is P4 (see FIG. 4), the determination result in step SB1 is “YES”, and it is determined whether at least one of the reference pixels in the region R is a valid pixel. (Step SB2). If the pixel of interest is P4, step SB2
Is "NO", the next target pixel is selected without changing the value of the target pixel (step SB3-
b, SB4). Further, the target pixel in the region R is P2, P
3 or P5 (see FIG. 4), the determination result in step SB1 is “YES”, the determination result in step SB2 is “YES”, and the target pixel is an invalid pixel regardless of its value. (White pixel) (step SB3-
a). Then, the next target pixel is selected (step S).
B4).

As a result of such a filtering process, as shown in FIG. 4, the connected effective pixel area is finally represented by one pixel P4. Thus, FIG.
Can be said to be a process in which the connected effective pixel area is represented by one effective pixel. Therefore, if the connected effective pixel area can be represented by one effective pixel, processing other than the filter processing shown in FIG. 5 can be adopted.

Referring again to FIG. 2, sampling processing is performed in a plurality of steps starting from step SA8.
Specifically, the binary image represented by the binary image data representing the result of the filter processing is divided into blocks including a plurality of pixels (step SA8), and the number of effective pixels in the block is determined for each block. Is greater than or equal to a predetermined number (step SA9), and the block whose determination result is “YES” is determined to be a valid (black) block (step S9).
A10-a), the block whose determination result is “NO” is regarded as an invalid (white) block (step SA10-a).
b). Then, the coarse binary image data having one block as one pixel is registered as the specific coarse binary image data (step SA11).

(1-2-2) Recognition Processing Next, recognition processing in the present method based on the above-described principle will be described with reference to FIG. However, the processing shown in this figure differs from the processing shown in FIG. 2 in that the input image is not a specific image but an input image to be recognized as a specific image, and that the binary image data after the sampling processing is different. Since the other processes are the same as those described above, the same processes as those in FIG. 2 are denoted by the same reference numerals, and description thereof will be omitted.

In the recognition processing in the present method, input coarse binary image data for the input image is obtained as a result of the sampling processing. Then, in step SA12, recognition processing of a specific image with respect to the input image is performed. Specifically, in the image represented by the input coarse binary image data,
A determination process is performed to determine whether an image represented by the registered specific coarse binary image data exists. This determination processing is performed by, for example, a general pattern matching method.

In this apparatus, not only a specific image,
The rotation image of the specific image is also a recognition target, and the above-described registration process is repeated by the number of images to be recognized to register a plurality of specific coarse binary image data. The presence or absence of a plurality of images represented by the value image data is determined. Of course, the binary image data after the binarization processing is rotated by the rotation angle of each image to be recognized to generate a plurality of binary image data, and the subsequent processing is performed on the binary image data. May be used to obtain a plurality of specific coarse binary image data.

(2) Configuration Next, the configuration of the present apparatus will be described. FIG. 6 is a block diagram showing the configuration of the present apparatus. The present apparatus shown in FIG.
A registration function for inputting specific image data representing a specific image to create and store specific coarse binary image data used in the recognition determination processing, and inputting input image data representing an input image to be recognized for the specific image And a recognition function of outputting a recognition result of the specific image with respect to the input image. Note that the image data input to the apparatus is stored in a predetermined color space (for example, RGB (red, green, blue) or YM
A series of color data having color information (full color information) in C (yellow, magenta, cyan), L ^* a ^* b ^* (any one color space selected from lightness, saturation, hue), etc. It is composed of Here, the following description will be made assuming that the input image data is composed of a series of RGB data.

In FIG. 6, reference numeral 1 denotes a binarizing unit for inputting image data, which performs a predetermined binarizing process on the input image data and outputs the processed image data. As shown in the figure, the binarizing unit 1 includes a binarizing unit A1a, a binarizing unit B1b,
It is composed of a binarizing section C1c and a binarizing section D1d.
By operating the binarizing unit A1a, binarizing unit B1b, binarizing unit C1c, and binarizing unit D1d in parallel, four types of binary image data are output for one image data.

The binarizing section A1a, the binarizing section B1b, the binarizing section C1c, and the binarizing section D1d perform a predetermined binarizing process on the input image data, and perform binarization. It outputs image data, and the binarization processing is set differently for each binarization unit. For example, 2
A binarization unit that performs binarization processing on RGB data and a binarization unit that binarizes YMC data correspond to image data in various data formats. You may be able to. In the present embodiment, each binarizing unit is set so as to be able to cope with a case where the partial image of the specific image has a plurality of colors (specific colors).

Here, an example of the configuration of the binarization unit A1a will be described. FIG. 7 is a diagram showing a configuration example of the binarization unit A1a. In this figure, the value of the R component in the input image data is 'R', the value of the G component is 'G', and the value of the B component is The value is described as 'B'. In FIG. 7, C1 is a comparator for R, which inputs 'R' and outputs '1' when this 'R' is equal to or more than a lower limit value 'Lr' of R and equal to or less than an upper limit value 'Ur' of R. , Outputs '0' in other cases. C2 is a comparator of G, which inputs "G", and when the "G" is equal to or more than the lower limit value "Lg" of G and equal to or less than the upper limit value "Ug" of G, "1" is used. '0' is output. C3 is a comparator of B, and inputs "B". If "B" is equal to or more than a lower limit value "Lb" of B and equal to or less than an upper limit value "Ub" of B, "1" is set. '0' is output.

GT1 is an AND gate, which performs an AND operation on the outputs from the respective comparators and outputs the result as binary image data. Therefore, the data output from the binarization unit A1a is output when the RGB components of the input data exist within the range defined by the upper limit value and the lower limit value (that is, when it can be determined that the color is a specific color). Is '1', otherwise it is '0'. That is, the configuration example shown in FIG.
This is a value conversion, and the following description will proceed based on this assumption.

As is apparent from the above description, in the configuration example shown in this figure, each upper limit value and each lower limit value,
By changing the contents of the AND gate GT1, 2
The contents of the valuation process can be changed. If the upper limit value and the lower limit value are matched, only the pixel of the specific color is “1”, and the pixel of a color slightly different from the specific color is “0”.
However, in the present embodiment, the upper limit value and the lower limit value are not the same in consideration of an error due to noise or the like.

The other binarization units also have a configuration similar to the above-described configuration, and are merely modified according to binarization conditions (including thresholds such as upper and lower limits). Absent. Since such a change is a matter to be appropriately designed, a specific description of the configurations of the binarization unit B1b, the binarization unit C1c, and the binarization unit D1d is omitted here.
Note that, in the present embodiment, 2 provided in the binarization section 1
Although the number of the value conversion units is “4”, it is needless to say that any number may be used.

In FIG. 6 again, reference numeral 2 denotes a binary image data selection unit, and at least one of the plurality of binary image data output from the binarization unit 1 is selected and selected. One binary image data based on the binary image data is output. The binary image data selection unit 2 has, for example, a configuration shown in FIG. In FIG. 8, GTa ~
GTd is an AND gate provided corresponding to each of the above-described binarization units. The output data of the binarization unit corresponding to GTd and “1” when selecting the binarization unit, GTd, etc. , A selection signal that becomes '0' is input, and AND
Take. GT2 is an AND gate AGTa, an AND gate BGTb, an AND gate CGTc, and an AND gate DGT.
An OR gate for ORing each output of d. That is,
The binary image data selecting unit 2 having the configuration shown in FIG. 8 outputs at least one of the output data of at least one binarizing unit in which the value of the corresponding selection signal is “1”. Is output when the output data exists, and is outputted as '0' in other cases. The binary image data selected by the binary image data selection unit 2 is binary image data output from a binarization unit that binarizes a specific color of a specific image to be recognized.

Referring again to FIG. 6, reference numeral 3 denotes a specific image portion determining section for determining whether or not a specific image has a partial image, and a binary image represented by the binary image data output from the binary image data selecting section 2. While scanning the top with a predetermined window area,
The above-described determination is performed by performing template matching on the image in the window area under a predetermined condition, and binary image data representing the determination result is output. A window area (see FIG. 13) used for template matching is set in advance in the specific image portion determination unit 3, and binary image data sufficient for setting the window area is set in the binary image data selection unit 2. Of line buffers (not shown) are provided to store the number of line buffers. Each line buffer is
Each has a capacity to store binary image data corresponding to one line of input image data, and the window area is set to a size that can accommodate a partial image of a specific image.

More specifically, the specific image portion determining section 3
If the result of the template matching satisfies a predetermined condition (that is, if it is determined that a partial image of the specific image exists), a pixel on the binary image corresponding to the central pixel of the partial image of the specific image (that is, The center pixel on the binary image is a valid pixel (value is “1”), and in other cases, the pixel is an invalid pixel (value is “0”). That is, in the binary image data output from the binary image data selection unit 2, when it is determined by template matching that a partial image of the specific image is present, the pixel located at the center of the window area is determined as an effective pixel. If it is not determined that a partial image of the image exists, the pixel located at the center of the window area is data representing a binary image in which the pixel is regarded as an invalid pixel.

Reference numeral 4 denotes a filtering unit which receives the binary image data output from the specific image part determining unit 3 and performs the following processing. Upon input of the binary image data from the specific image portion determination unit 3, the filtering unit 4 determines whether the connected valid pixel region exists in the binary image represented by the binary image data. The region is represented by one effective pixel, and binary image data representing a binary image composed of effective pixels that are not connected to each other is output. For example, as shown in FIG. 9, when binary image data representing the input image 8a to the filtering unit is input from the specific image portion determination unit 3 to the filtering unit 4, the filtering unit 4 outputs It outputs binary image data representing the image 8b. The reason why the filtering unit 4 is provided is that there is a possibility that a pixel adjacent to an effective pixel may also be determined as an effective pixel depending on the conditions set in the specific image part determination unit 3. 1 consisting of multiple effective pixels
This is to avoid a situation in which one connected effective pixel area extends over two blocks, and eventually the probability of non-detection increases.

The sampling section 5 includes a filtering section 4
The binary image represented by the binary image data output from is divided into a plurality of blocks, the presence or absence of an effective pixel is checked for each block, and an effective block / invalid block is determined based on the result, and one block is determined. Is output as binary image data. Specifically, the sampling unit 5 sets 4 × 4 pixels on the binary image represented by the binary image data output from the filtering unit 4 as one block, and at least one valid pixel ( When there is a black pixel), the block is set as an effective block (black block), otherwise, the block is set as an invalid block (white block), and binary image data with one block as one pixel is output. . For example, as shown in FIG. 10, 2 representing the output image 8b (16 × 16 pixels) from the filtering unit
When the value image data is input to the sampling unit 5,
The sampling unit 5 outputs an image 9 from the sampling unit.
Binary image data representing (4 × 4 pixels) is output. That is, the sampling unit 5 operates so as to reduce the resolution of the binary image while maintaining the approximate position information of each effective pixel. However, the sampling unit 5 switches the output destination of the binary image data between the reference pattern storage unit 6 and the specific image determination result output unit 7 according to an instruction from a control device (not shown). Specifically, the sampling unit 5 sets the reference pattern storage unit 6 as the output destination of the binary image data when performing the registration process and the specific image determination result output unit 7 as performing the recognition process.

The reference pattern storage section 6 stores the binary image data output from the sampling section 5 as a reference pattern used for the recognition judgment processing in the specific image judgment result output section 7, and stores each of the identification targets. It has a storage capacity capable of storing a reference pattern for an image. The specific image determination result output unit 7 outputs the second image output from the sampling unit 5.
A recognition determination process is performed to determine whether or not the reference pattern stored in the reference pattern storage unit 6 exists in the value image data, and to output a determination result as a recognition result. In the present embodiment, a general pattern matching method is adopted as the method of the recognition determination processing. In addition, when the determination result is a result indicating that the pattern exists, the specific image determination result output unit 7 outputs the determination result and information regarding the pattern as a recognition result.

(3) Operation Next, the operation of the present apparatus having the above configuration will be described. (3-1) Premise Here, the input image is the input image 10 shown in FIG. 11, the data format of the input image data is RGB data, and the specific image to be recognized is the specific image A11a in FIG.
It is assumed that As shown in FIG. 12, the specific image A1
1a is a circular partial image A12a, a partial image B12b,
Partial image C12c, partial image D12d, partial image E12
e, and a partial image F12f. The shape of each partial image is a filled circle, and its color is
12a and partial image B12b, partial image C12c and partial image D12d, and partial image E12e and partial image F12f have the same specific color, respectively. Also, here, as shown in FIG. 11, a specific image B11b, a specific image C11c, and a specific image D11d obtained by rotating the specific image A11a by 90 degrees, 180 degrees, and 270 degrees in the clockwise direction in the figure.
Is also a recognition target image. As is clear from FIG. 11, the input image 10 includes a specific image A11a, a specific image B11b, a specific image C11c, and a specific image D11d.

Before inputting an input image, it is assumed that a condition (here, a threshold value) for performing binarization using a specific color of a specific image is set in the binarization unit 1 of FIG. .
Specifically, the binarization unit A1a includes R, R corresponding to a specific color common to the partial image A12a and the partial image B12b.
Upper and lower limit values of G and B, binarization section B1b
The upper limit value and the lower limit value of each of R, G, and B corresponding to the specific color common to the partial image C12c and the partial image D12d, and the binarizing unit C1c sets the specific color common to the partial image E12e and the partial image F12f. Corresponding R, G, B
It is assumed that respective upper and lower limits are set. Note that the upper limit and the lower limit have a sufficient width so that each specific color can be reliably extracted in consideration of printing accuracy of a document on which an input image is printed, accuracy of a reading device, and the like. Should be set.

Each selection signal (see FIG. 8) supplied from the control device (not shown) to the binary image data selection section 2 is
It is assumed that the setting is made so that only the output data of the binarization unit A1a, the binarization unit B1b, and the binarization unit C1c are selected.
That is, the value of each selection signal of the binarization unit A1a, the binarization unit B1b, and the binarization unit C1c is set to "1", and the value of the selection signal of the binarization unit D1d is set to "0". . By doing so, the output data from the binary image data selection unit 2 becomes binary image data obtained by extracting a plurality of specific color components forming a specific image from the input image data.

Further, the specific image portion judging section 3 has
It is assumed that an 11 × 11 window area 13 shown in FIG. The window area 13 includes mask areas 14a and 1a for recognizing a partial image of the specific image.
4b is set so that each center coincides with the center of the window area 13.

(3-2) Specific Operation First, when input image data representing an input image shown in the specific image determination result output unit 7 is input to the binarization unit 1, The input image data is binarized in a binarizing unit,
It is output as four binary image data. These four binary image data are input to the binary image data selector 2. The binary image data selector 2 outputs the output data of the binarizer A1a, the binarizer B1b, and the binarizer C1c. Since only binary image data is set to be selected, the binary image data output from the binary image data selection unit 2 is data obtained by extracting a plurality of specific color components forming a specific image from the input image data.

The binary image data output from the binary image data selection unit 2
The value image data is input to the specific image portion determination unit 3, where the image represented by the binary image data is 11 ×
Scanning by the eleven window areas 13 is performed. In this scanning, for the image in the window area 13,
Mask area A1 set in advance on window area 13
4a and template matching using the mask area B14b. Then, the number of black pixels in the mask area A14a is equal to or more than a predetermined number and the mask area B1
Only when the number of black pixels in 4b is equal to or less than a predetermined number,
Output 2 prepared separately from the binary image data to be scanned
In the value image data, data representing a black pixel is written at a data position corresponding to the center of the window area 13. That is, in the binary image represented by the output binary image data, the pixel at the position corresponding to the center of the window area 13 is a black pixel. On the other hand, if the number of black pixels in the mask area A14a is less than the predetermined number or the number of black pixels in the mask area B14b exceeds the predetermined number, the binary image data for output includes Data representing a white pixel is written at a data position corresponding to the center. That is, in the binary image represented by the binary image data for output, the pixel at the position corresponding to the center of the window area 13 is a white pixel.

Here, a specific processing result will be described with reference to the example shown in FIG. However, the mask area A1
When the number of black pixels in 4a is 30 or more and the number of black pixels in the mask area B14b is 0 or less, data representing a black pixel is output. In other cases, data representing a white pixel is output. In the value image data, it is assumed that the specific image portion determination unit 3 is set to write the data at a data position corresponding to the center of the window area 13.

As shown in FIG. 14, the binary image represented by the binary image data from the binary image data selection unit 2 includes the input image A15a of the specific image part determination unit, and the specific image part determination is performed. When the above-described processing is performed on the input image A15a of the unit in the specific image part determination unit 3, the binary image represented by the binary image data for output includes the output image A16a of the specific image part determination unit become. That is,
For a partial image of one specific image, partial image candidates are extracted only twice. On the other hand, the binary image represented by the binary image data from the binary image data selection unit 2 includes the input image B15b of the specific image part determination unit. When the above-described processing is performed in the specific image part determination unit 3, the binary image represented by the binary image data for output includes the output image B16b of the specific image part determination unit. That is, a partial image candidate is extracted only once for a partial image of one specific image. The output binary image data created in this way is output to the filtering unit 4.

Here, the reason for performing the above-described template matching will be described. For example, when the present apparatus is applied to a scanner or a copying machine, variations in color and shape occur in a partial image of a specific image in an input image due to the printing accuracy of a document, the accuracy of a reading device, and the like. If pattern matching is simply performed without considering this variation, the image is not determined to be a partial image even though it is a partial image of the specific image. There is a possibility that “non-detection” that is not recognized as a specific image may occur. Therefore, in the present apparatus, in consideration of the variation, if the difference from the original partial image is within a predetermined range, that is, if the number of black pixels in the mask area A14a is equal to or more than the predetermined number and the mask area B14b When the number of black pixels is equal to or smaller than a predetermined number, the pixel located at the center of the window area 13 is set as a black pixel and is left as a partial image candidate.

The binary image data output from the specific image portion determination unit 3 is input to the filtering unit 4, where
Filtering processing is performed. The filtering process is a process in which, among black pixels in the binary image represented by the binary image data, a plurality of adjacent black pixels (connected effective pixels) are represented by one black pixel. That is, this is a process of narrowing down a plurality of candidates obtained for one partial image to one candidate. As a result, the filtering unit 4
The number of black pixels in the binary image represented by the binary image data output from is equal to the number of partial images of the specific image in the document image.

For example, when the binary image represented by the binary image data from the specific image portion determination unit 3 includes the input image 8a to the filtering unit in FIG. 9, the filtering process for the input image 8a to the filtering unit is , According to the algorithm shown in FIG. That is, the input image 8a to the filtering unit is determined by the reference area shown in FIG.
Is scanned from the upper right to the lower left in the figure, and only the black pixel located at the lower left end among the plurality of adjacent black pixels is left. As a result, binary image data representing the output image 8b from the filtering unit in FIG. 9 is obtained as a filtering result for the input image 8a to the filtering unit.
The reason why the partial image candidates are narrowed down by the filtering process will be described later in detail, but it is necessary to eliminate the influence of allowing the variation in the specific image portion determining unit 3 in the preceding stage of the sampling process. .

The binary image data output from the filtering unit 4 is subjected to a sampling process here. The sampling process is a process of lowering the resolution of the image while maintaining the characteristics of the image. The characteristics of the image to be maintained here are the binary images represented by the binary image data from the filtering unit 4. Is position information of each black pixel. Specifically, in the sampling unit 5, the filtering unit 4
The binary image represented by the binary image data is divided into blocks each including 4 × 4 pixels, and a block in which at least one black pixel (effective pixel) exists in its own block is a black block (effective block) The other blocks are set as white blocks (invalid blocks), and coarse binary image data having one block as one pixel is finally output.

For example, when the binary image represented by the binary image data from the filtering unit 4 includes the output image 8b from the filtering unit in FIG. 9, the sampling unit 5 obtains the output image 9 shown in FIG. . That is, as a result of the sampling process on the output image 8b from the filtering unit, binary image data representing the output image 9 is output from the sampling unit 5. By performing the sampling processing in the filtering unit 4, in the present apparatus, the black pixels corresponding to one partial image in the binary image represented by the binary image data from the filtering unit 4 are spread over a plurality of blocks. There is no. That is, for example, the approximate position of the black pixel in the output image 8b from the filtering unit can be known from the position of the black block in the output image 9 from the sampling unit. Therefore, although the accuracy is reduced, the position information of each black pixel is maintained in the coarse binary image data output from the sampling unit 5. This is an advantage of performing the filtering process.

The coarse binary image data output from the sampling unit 5 is input to the specific image determination result output unit 7, where a pattern matching process using a preset reference pattern is performed on the coarse binary image data. Done. Specifically, the specific image determination result output unit 7 scans the binary image represented by the coarse binary image data by using a window area having a size (for example, 4 × 4 pixels) including the reference pattern, and The same pattern as the reference pattern in the window area (pattern consisting of black and white pixels)
Is determined, and a determination result is output.
In this apparatus, since a general pattern matching method is employed, a detailed operation in the pattern matching processing will not be described.

The target of pattern matching in the specific image determination result output unit 7 is the specific image A11a.
And a specific image B11b, a specific image C11c, and a specific image D11d, which are rotated images thereof. If a general pattern matching method is adopted, four types of reference patterns are required, and the number of comparisons in the pattern matching process is quadrupled. . However, when compared with the conventional image recognition method that does not perform the preceding process, the number of comparisons required for one pattern matching is reduced from 16 × 16 = 256 [times] to 4 × 4 = 16 [times]. Thus, even if the number of comparisons is quadrupled, the total number of comparisons (16 × 4 = 64)
[Times]) is a sufficiently small number of times as compared with the conventional image recognition method that does not support rotation. Of course, if an image recognition method with a smaller number of comparisons is adopted, the number of comparisons can be further reduced.

When the same pattern as the reference pattern is detected, the judgment result output from the specific image judgment result output unit 7 becomes information indicating that the specific image has been recognized.
If the pattern cannot be detected, the information indicates that the specific image has not been recognized. The specific image determination result output unit 7 outputs information indicating the type (rotation angle) of the matched reference pattern and information indicating the scanning position. If these information are referred to when the pattern is detected, , Which pattern is detected at which position. Of course, if the device or user using the output from the specific image determination result output unit 7 needs only information indicating recognition / non-recognition of the specific image, the specific image determination result output unit 7 outputs another information. No information needs to be output.

As described above, according to the present apparatus, a partial image of a specific image is detected at a normal resolution, and an image obtained by lowering the resolution while retaining the position information of the detected partial image with a certain degree of accuracy. , The pattern matching is performed, so that sufficient detection accuracy can be achieved.
As a result, even when not only the specific image but also the rotated image is recognized, sufficient detection accuracy can be achieved while significantly reducing the circuit scale and labor required for pattern matching.

(4) Modification Next, a modification of the present apparatus will be described. FIG. 15 is a block diagram showing a basic configuration of the present modification. The configuration shown in FIG. 15 differs from that shown in FIG. 6 in that the filtering unit 4 and the sampling unit 5 are replaced, that is, the filter The point is that the order of the processing and the sampling processing is reversed. Even with such a configuration, the same effect as that of the present device can be obtained. Furthermore, by adopting such a configuration, even if the size of the image data input to the binarization unit 1 is the same, the size of the image data input to the filtering unit 4 is reduced by the sampling unit 5, so that the filtering process is performed. In this case, the required memory capacity can be reduced. Therefore, the circuit scale can be reduced as compared with the present apparatus having the configuration of FIG.

(5) Supplement Although it depends on the resolution of the specific image and the input image read by the reading device and the content of the specific image, in the above-described device, the position of the partial image of the specific image is changed due to some error. May correspond to a block different from the original block (see FIG. 10). Therefore, a reference pattern may be created with a margin in consideration of such a deviation. For example, instead of using the coarse binary image data after the sampling of the specific image as it is for pattern matching, the deviation of the position of each partial image is considered, and as shown in FIG. Pattern matching may be performed after white pixels (white blocks) within a predetermined range around the area are also determined as areas where black pixels are present (areas indicated by oblique lines). However, if at least one black pixel exists in the region, it is determined that a black pixel exists for the region. If a black pixel exists for the entire region and no white pixel exists for other regions, the specific image is determined. It is determined that it has been recognized. As is clear from FIG. 16, the predetermined range can be set for each black pixel.
Conversely, even if a certain degree of error occurs, the contents of the specific image and the black pixels (see the output image 8b from the filtering unit in FIG. 9) indicating the position of the partial image of the specific image are not positioned near the boundary of the block. The size of the block may be set.

Further, in the above-described apparatus, the partial image is recognized using the window area 13 and the mask areas 14a and 14b shown in FIG.
Any angle can be used as long as a reference pattern corresponding to a specific image rotated at various angles other than 0, 180, and 270 degrees is prepared. Moreover, since the reference pattern is represented by coarse binary image data having a small data amount, there is an advantage that a large number of reference patterns can be held with a small memory capacity as compared with a case where the binary image data is held as it is. is there. In other words, the feature that the shape of the partial image does not change even when the specific image is rotated makes it possible to cope with the rotation of the image while suppressing an increase in the processing amount and the circuit scale.

In this embodiment, the mask region 14a
Although a circular figure is adopted as the elements 14b and 14b, a figure in which a rotated image such as a regular polygon completely overlaps itself may be adopted. However, the rotation angle of the image that can be handled differs depending on the figure used. For example, when a square is adopted, only rotations of 0 degrees, 90 degrees, 180 degrees, and 270 degrees can be supported.
Only rotations of 20 degrees and 240 degrees are possible. In addition,
The angle is a clockwise rotation angle.

By the way, in the present apparatus, if a specific image exists in the window area during scanning for performing pattern matching, the number of black pixels in the window area becomes equal to the number of black pixels in the reference pattern. By utilizing this fact, pattern matching may be performed only when the number of black pixels in the window area matches the number of black pixels in the reference pattern, so that the processing may be simplified and the accuracy may be improved.

The reference pattern may be obtained and set by an arbitrary method other than the method described above. Further, when the specific color of the specific image to be recognized is one color, in the configuration shown in FIG. 6, the binarizing unit 1 is composed of one binarizing unit, and the binary image data selecting unit 2 is deleted. May be. Further, in the sampling process for the input image, binary image data may be obtained by using one block as a plurality of pixels. That is, the degree of simplification of the data by the sampling process can be arbitrarily set.
Needless to say, it is necessary to simplify the reference pattern for the specific image to the same extent as described above.

Note that the image recognition methods according to the above-described embodiments and modifications are realized as software described in a programming language. Further, the image recognition device according to the above-described embodiment and the modified example can be realized by a computer system and software.

[0072]

As described above, according to the present invention,
A specific image can be easily and quickly recognized from an input image. In addition, not only the specific image but also the rotated image can be recognized at high speed. Therefore, the specific image can be recognized at high speed with a simple configuration and processing.

[Brief description of the drawings]

FIG. 1 is a flowchart showing a flow of a recognition process in an image recognition method realized by the apparatus.

FIG. 2 is a flowchart illustrating a flow of a registration process in an image recognition method realized by the image recognition device according to the first embodiment of the present invention.

FIG. 3 is a diagram showing a region R used in a filtering process by a filtering unit 4 of the apparatus.

FIG. 4 is a diagram for explaining the filter processing.

FIG. 5 is a flowchart showing a flow of the filter processing.

FIG. 6 is a block diagram showing a configuration of the same device.

FIG. 7 shows a binarization unit A1 included in the binarization unit 1 of the apparatus.
It is a figure showing the example of composition of a.

FIG. 8 is a diagram showing a configuration example of a binary image data selection unit 2 of the apparatus.

FIG. 9 is a diagram for explaining processing of a filtering unit 4 of the apparatus.

FIG. 10 is a diagram for explaining processing of a sampling unit 5 of the apparatus.

FIG. 11 is a diagram illustrating an example of a document image including a specific image.

FIG. 12 is a diagram illustrating an example of a specific image.

FIG. 13 is a diagram for explaining template matching processing by a specific image portion determination unit 3 of the same device.

FIG. 14 is a diagram for explaining processing of a specific image portion determination unit 3 of the same device.

FIG. 15 is a block diagram showing a configuration of a modification of the apparatus.

FIG. 16 is a diagram illustrating a recognition determination process performed by a specific image determination result output unit 7 of the apparatus.

[Explanation of symbols]

1 binarization section, 1a binarization section A, 1b binarization section B,
1c Binarization section C, 1d Binarization section D, 2 ... Binary image data selection section, 3 ... Specific image portion determination section, 4 ... Filtering section, 5 ... Sampling section, 6 ... Reference pattern storage section, 7 ... Specific image determination result output unit 10 ... Input image, 1
1a specific image A, 11b specific image B, 11c specific image C, 11d specific image D, 12a partial image A, 1
2b partial image B, 12c partial image C, 12d partial image D, 12e partial image E, 12f partial image F, 1
4a: Mask area A, 14b: Mask area B, 15a ...
Input images A and 15b of the specific image portion determining section: input images B and 16a of the specific image portion determining section, output images A and 16b of the specific image portion determining section and output images B of the specific image portion determining section

──────────────────────────────────────────────────続 き Continuation of the front page (56) References JP-A-59-191678 (JP, A) JP-A-3-160576 (JP, A) JP-A-63-211474 (JP, A) JP-A 8- 137908 (JP, A) (58) Field surveyed (Int. Cl. ⁷ , DB name) G06T ⁷ /00-7/60 G06T 1/00 JICST file (JOIS)

Claims (20)

(57) [Claims] 1. An image recognition method for recognizing a specific image composed of a plurality of partial images having the same shape and the same size from an input image, comprising: recognizing the partial image from the input image; An input image part determining step of generating a binary image in which pixels of the input image are valid pixels and pixels of a position where the partial image does not exist are invalid pixels; and a plurality of the binary images generated in the input image part determining step. An input image sampling step of dividing the block including the effective pixels into a binary image in which the blocks including the effective pixels are expressed as an effective block and the blocks not including the effective pixels as an invalid block; A reference pattern which is obtained by simplifying the binary image and the specific image simplified in the sampling step in units of the blocks. And an overall determination step of outputting the result of the matching. 2. The method according to claim 1, wherein in the input image sampling step, the second image generated in the input image portion determination step is generated.
The value image is divided by a block having a size including a plurality of pixels, a block including the effective pixels is expressed as an effective block, a block not including the effective pixels is expressed as an invalid block, and one effective block is expressed as one effective block. A pixel, the one invalid block is represented by one invalid pixel, and a binary image is generated. In the overall determination step, the binary image and the specific image generated in the input image sampling step are previously stored in the block unit. Then, each block is matched with a reference pattern represented by one pixel.
The image recognition method according to claim 1, wherein a result of the matching is output. 3. When a plurality of effective pixels are adjacent to each other on the binary image generated in the input image portion determining step, the other valid pixels except one effective pixel are regarded as invalid pixels. Having an input image filtering step of changing the binary image changed in the input image filtering step,
2. The image processing apparatus according to claim 1, wherein the image data is divided into blocks each having a size including a plurality of pixels, and a block including the effective pixels is simplified as a valid image, and a block not including the effective pixels is expressed as an invalid block. The image recognition method described in 1. 4. When a plurality of effective pixels are adjacent to each other on the binary image simplified in the input image sampling step, another effective pixel excluding one effective pixel is regarded as an invalid pixel. In the overall determination step, the binary image generated in the input image filtering step is matched with a reference pattern that is a simplified representation of the specific image in block units. 2. The image recognition method according to claim 1, wherein a result of the matching is output. 5. An input image binarizing step of binarizing the input image based on a preset condition to generate a binary image, wherein the input image portion determining step includes: Recognizing the partial image from the binary image generated in the step, a pixel at a position where the partial image exists as an effective pixel,
A pixel at a position where the partial image does not exist is regarded as an invalid pixel 2
2. The image recognition method according to claim 1, wherein a value image is generated. 6. A method of recognizing the partial image from the specific image and generating a binary image in which a pixel at a position where the partial image exists is an effective pixel and a pixel at a position where the partial image does not exist is an invalid pixel. The binary image generated in the image part determining step and the specific image part determining step is divided by the block, and a block including the effective pixel is represented as a valid block, and a block not including the valid pixel is represented as an invalid block. And a specific image sampling step of generating a reference pattern that is a binary image obtained by the above. In the overall determination step, the binary image simplified in the input image sampling step and the specific image generated in the specific image sampling step are provided. 2. The method according to claim 1, wherein the matching is performed with the reference pattern, and a result of the matching is output. Image recognition method. 7. When a plurality of valid pixels are adjacent to each other on the binary image generated in the input image portion determining step, the binary image is regarded as an invalid pixel except for one valid pixel. Input image filtering step of changing, and when a plurality of effective pixels are adjacent on the binary image generated in the specific image portion determination step, other effective pixels except one effective pixel are regarded as invalid pixels. A specific image filtering step of changing the binary image, wherein in the input image sampling step, the binary image changed in the input image filtering step is
Dividing by a block having a size including a plurality of pixels, simplifying a block including the effective pixels into a binary image expressed as an effective block, and a block not including the effective pixels as an invalid block, in the specific image sampling step, The binary image generated in the specific image filtering step;
7. The reference pattern according to claim 6, wherein the reference pattern is a binary image that is divided by the block, and a block including the effective pixel is expressed as an effective block, and a block not including the effective pixel is expressed as an invalid block. Image recognition method. 8. When a plurality of valid pixels are adjacent to each other on the binary image simplified in the input image sampling step, the other valid pixels except one valid pixel are regarded as invalid pixels. An input image filtering step of changing the reference pattern, and when a plurality of effective pixels are adjacent to each other on the reference pattern generated in the specific image sampling step, the other effective pixels except one effective pixel are regarded as invalid pixels and the reference A specific image filtering step of changing a pattern, wherein in the overall determination step, the binary image generated in the input image filtering step is matched with the reference pattern changed in the specific image filtering step, The image recognition method according to claim 6, wherein a result of the matching is output. 9. An input image binarizing step of binarizing the input image based on a preset condition to generate a binary image, and binarizing the specific image based on a color of the partial image. A specific image binarizing step of generating a binary image, wherein the input image part determining step recognizes the partial image from the binary image generated in the input image binarizing step, and The pixel at the position where the image exists is the effective pixel,
A pixel at a position where the partial image does not exist is regarded as an invalid pixel 2
Generating a value image, the specific image part determination step recognizes the partial image from the binary image generated in the specific image binarization step, and sets a pixel at a position where the partial image exists as an effective pixel;
A pixel at a position where the partial image does not exist is regarded as an invalid pixel 2
The image recognition method according to claim 6, wherein a value image is generated. 10. The image recognition method according to claim 1, wherein the shape of the partial image is such that the rotated image completely overlaps itself. 11. An image recognition apparatus for recognizing a specific image composed of a plurality of partial images having the same shape and the same size from an input image, the apparatus recognizing the partial image from an image represented by input image data, The pixel at the position where the partial image exists is an effective pixel,
A pixel at a position where the partial image does not exist is regarded as an invalid pixel 2
A partial determining unit that generates binary image data representing a value image; and dividing the binary image represented by the binary image data generated by the partial determining unit into blocks each having a size including a plurality of pixels. A sampling unit that simplifies a block including the valid pixels into a valid image and a block that does not include the valid pixels into a binary image, and generates binary image data representing the binary image; Storage means for storing reference pattern data representing a reference pattern in which an image is simplified in block units in advance; and a binary image represented by the binary image data generated by the sampling means and the storage means. An overall format that matches the reference pattern represented by the stored reference pattern data and outputs the result of the matching And means, the image data, wherein the portions determining means for inputting the image recognition apparatus, characterized in that the image data representing the input image. 12. The block including the effective pixel, wherein the sampling unit divides a binary image represented by the binary image data generated by the partial determination unit into blocks each having a size including a plurality of pixels. Is an effective block,
A block not including the effective pixel is represented as an invalid block, and one effective block is simplified into a binary image represented by one effective pixel and one invalid block by one invalid pixel. The storage means stores reference pattern data representing a reference pattern in which each block is represented by one pixel after the specific image is simplified and expressed in advance in the block unit. The image recognition device according to claim 11, wherein: 13. When a plurality of effective pixels are adjacent to each other on the binary image represented by the binary image data generated by the partial determination unit, another effective pixel excluding one effective pixel is removed. Filtering means for changing the binary image data as invalid pixels, wherein the sampling means converts the binary image represented by the binary image data changed by the filtering means to a size including a plurality of pixels. And simplifies the block including the effective pixel into a binary image expressing the block including the effective pixel as an effective block and the block including the effective pixel as an invalid block, and generates binary image data representing the binary image. The image recognition device according to claim 11, wherein: 14. When a plurality of valid pixels are adjacent to each other on the binary image represented by the binary image data generated by the sampling means, the other valid pixels except one valid pixel are invalidated. A filtering unit that changes the binary image data as a pixel, wherein the overall determination unit includes a binary image represented by the binary image data generated by the filtering unit;
The image recognition apparatus according to claim 11, wherein matching is performed with the reference pattern represented by the reference pattern data stored in the storage unit, and a result of the matching is output. 15. Binarization means for binarizing input image data based on preset conditions to generate binary image data, wherein said partial determination means is generated by said binarization means. The binary image data is input, the partial image is recognized from the binary image represented by the binary image data, a pixel at a position where the partial image exists is a valid pixel, a position at which the partial image does not exist The image recognition apparatus according to claim 11, wherein binary image data representing a binary image in which the pixels of (b) are invalid pixels is generated. 16. A binarizing means for performing binarization on input image data independently for each condition based on a plurality of preset conditions to generate a plurality of binary image data. And selecting means for generating one piece of binary image data based on the plurality of pieces of binary image data generated by the binarizing means. Binary image data is input, the partial image is recognized from the binary image represented by the binary image data, the pixel at the position where the partial image exists is the effective pixel, the pixel at the position where the partial image does not exist The image recognition apparatus according to claim 11, wherein binary image data representing a binary image in which is set as an invalid pixel is generated. 17. The image data input by the partial determination unit is the input image and the specific image, and the storage unit is configured to output the specific image if the image data input by the partial determination unit is the specific image. 12. The method according to claim 11, wherein the binary image data generated by the sampling unit is stored as the reference pattern data.
An image recognition device according to claim 1. 18. The image data input by the partial determination unit is image data representing the input image and image data representing the specific image. The storage unit stores the image data input by the partial determination unit. 15. The image recognition apparatus according to claim 13 , wherein in the case of image data representing the specific image, the binary image data generated by the filtering unit is stored as the reference pattern data. 19. The image recognition apparatus according to claim 11, wherein the shape of the partial image is such that the rotated image completely overlaps itself. 20. A program for implementing an image recognition method for recognizing a specific image composed of a plurality of partial images having the same shape and the same size from an input image, the program recognizing the partial image from the input image, and An input image portion determining step of generating a binary image in which a pixel at a position at which a partial image exists is an effective pixel, and a pixel at a position at which the partial image does not exist as an invalid pixel; An input image sampling step of dividing the image into blocks having a size including a plurality of pixels, and simplifying the block including the effective pixels into a binary image in which the blocks including the effective pixels are represented as valid blocks and the blocks not including the valid pixels are represented as invalid blocks; The binary image and the specific image simplified in the input image sampling step are simplified in block units in advance. A program for realizing an overall determination step of matching with a reference pattern expressed as a result and outputting the result of the matching.