JPH11316837A - Image processing method for pattern matching and image processor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain an image processing method which performs appropriate pattern matching to an image undergoing discrete cosine transformation(DCT) by comparing detected similarity with preliminarily defined threshold and deciding whether or not a pattern resembling to a reference pattern is included in an input DCT image according to the comparison result. SOLUTION: A reference DCT pattern producing means 1 performs DCT of a reference pattern and produces a reference DCT pattern. A comparison area setting means 2 sets a comparison area that is the same size as the reference pattern from an input DCT image. A similarity detecting means 3 detects the similarity between the reference DCT pattern and the comparison area. A similarity deciding means 5 compares similarity detected by the means 3 with a predetermined threshold, and when similarity that is larger than the threshold exists, it is decided that a pattern resembling to the reference pattern is included in the input DCT image.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image processing method and an image processing method for pattern matching which can search whether or not a pattern similar to a reference pattern is included in a compressed image using discrete cosine transform. It concerns the device.

[0002]

2. Description of the Related Art FIG. 15 is a flow chart showing a conventional image processing method for pattern matching shown in, for example, Pattern Understanding (Yoshiaki Shirai, Ohmsha), and FIG. 16 is a conventional image processing method for pattern matching. FIG. 17 is an explanatory diagram showing a reference pattern in the method, and FIG. 17 is an explanatory diagram showing an input image in a conventional image processing method for pattern matching. In the figure, 101 is a reference pattern,
102 is the size of the reference pattern 101 (16 × 16),
103 is an input image, 104 is a first comparison area set in the input image 103, and 105 is a second comparison area set in the input image 103. ST101 is a comparison area setting step, ST102 is a dissimilarity detection step, ST
103 is a repetition determination step, and ST104 is a dissimilarity determination step.

Next, the operation will be described. First, a part of the input image 103 is set as a region to be compared with the reference pattern 101 (step ST101). As shown in FIG. 16, when the size of the reference pattern 101 is 16 × 16,
As shown in FIG.
Is set in the first comparison area 104. Next, the degree of dissimilarity between the first comparison area 104 and the reference pattern 101 is detected (step ST102). The dissimilarity is calculated, for example, as the sum of absolute differences between the values of the pixels. Next, it is determined whether or not all of the predetermined areas have been compared (step ST10).
3) If no comparison is made, the process returns to step ST101, and the position is shifted from the first comparison area 104 as the previous area to the second comparison area 105 as a new area as shown in FIG. cure. Next, all detected dissimilarities are compared with a predetermined threshold value. If there is a degree of dissimilarity smaller than the threshold, it is determined that a pattern similar to the reference pattern is included in the input image (step ST104). On the other hand, if all the dissimilarities are equal to or larger than the threshold value, it is determined that the input image 103 does not include a pattern similar to the reference pattern.

[0004]

Since the conventional image processing method and image processing apparatus for pattern matching are configured as described above, the input image is converted to a discrete cosine transform (DiscreteCosine Transform).
In the case of an image that has been subjected to (rmation: DCT), it is necessary to perform inverse DCT before performing the matching process, and there has been a problem that the amount of calculation of the entire process increases.
Since DCT is performed in all standard image compression such as JPEG and MPEG, this problem is serious in performing a large-scale image search for a database of compressed images.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned problems, and an image processing method and an image processing method for pattern matching that can perform appropriate pattern matching on an image subjected to DCT. The aim is to obtain a device.

[0006]

An image processing method for pattern matching according to the present invention performs a discrete cosine transform on a reference pattern to create a reference DCT pattern, and has the same size as the reference pattern from an input DCT image. The comparison area is set, the similarity between the reference DCT pattern and the comparison area is detected from the inner product using the DCT coefficient, the normalized correlation, and the like, and the same size area as all the reference patterns whose positions are changed in the input image is detected. It is determined whether or not it has been compared with all the areas having the same, and the detected similarity is compared with a predetermined threshold. If a similarity greater than this threshold exists, the input DCT image contains When it is determined that a pattern similar to the reference pattern is included, and when all the similarities are equal to or less than the threshold, a pattern similar to the reference pattern is included in the input DCT image. In which was to determine the rare no.

An image processing method for pattern matching according to the present invention performs a discrete cosine transform on a reference pattern to generate a reference DCT pattern, and a plurality of input DCT patterns.
One of the T images is selected, a comparison area of the same size as the reference pattern is set from the one input DCT image selected in this input image selection step, and the reference DCT is determined from inner products using DCT coefficients, normalized correlations, and the like. The similarity between the pattern and the comparison region is detected, and it is determined whether or not all the reference patterns whose positions have been changed in the input image are compared with all regions having the same size region. The determined thresholds are compared, and if a similarity greater than this threshold exists, it is determined that a pattern similar to the reference pattern is included in the input DCT image, and all the similarities are thresholded. If the value is equal to or less than the value, it is determined that the input DCT image does not include a pattern similar to the reference pattern, and it is determined whether all the input DCT images have been selected. If not-option is the last back to the input image selection step is obtained as reselect another input DCT image.

In the image processing method for pattern matching according to the present invention, in the step of creating a reference DCT pattern, the X-direction shift and the Y-direction shift of the reference point of the area where the discrete cosine transform is performed and the reference point of the reference pattern are performed. In the step of creating a plurality of reference DCT patterns while changing the displacement and detecting the similarity, the plurality of reference DCT patterns are used.
The similarity between the T pattern and the comparison area is detected.

An image processing method for pattern matching according to the present invention has a coefficient selection step of selecting a DCT coefficient used for similarity detection, and the similarity detection step includes only a selected part of the DCT coefficients. Is used to detect the similarity.

[0010] In the image processing method for pattern matching according to the present invention, between the step of selecting a coefficient and the step of creating a reference DCT pattern, the calculation accuracy in similarity detection is set according to the selected DCT coefficient. It is like that.

According to the image processing method for pattern matching according to the present invention, in the step of setting the calculation accuracy,
The calculation accuracy in the similarity detection is set according to the frequency of the selected DCT coefficient.

An image processing method for pattern matching according to the present invention is such that in the step of creating a reference DCT pattern, discrete cosine transform is performed on the reference pattern to create only necessary DCT coefficients of the reference DCT pattern. It is.

An image processing method for pattern matching according to the present invention performs a discrete cosine transform on a reference pattern to create a reference DCT pattern, selects a DCT coefficient from a low frequency range, and outputs a reference pattern from an input DCT image. A comparison area having the same size as that of the reference area is set, the similarity between the reference DCT pattern and the comparison area is detected using only selected DCT coefficients, and it is determined whether or not the similarity is compared with all the predetermined areas. , If not compared to all regions, the first
Returning to the comparison area setting step, the comparison area is set again by shifting the position from the previous time, the candidate area is determined based on the first similarity, the DCT coefficient is selected from the high frequency area, and the comparison is performed from among the candidate areas. A region is selected, the similarity between the reference DCT pattern and the comparison region is detected using only some of the selected DCT coefficients, and it is determined whether or not all regions have been compared with the candidate region. If not, the process returns to the step of setting the second comparison area, selects a candidate area different from the previous area, compares the detected similarity with a predetermined threshold, and sets a similarity larger than the threshold. If there is a degree, it is determined that a pattern similar to the reference pattern is included in the input image.If all the degrees of similarity are equal to or less than the threshold, the input image includes a pattern similar to the reference pattern. Judge not to be In which was to so that.

An image processing method for pattern matching according to the present invention includes the steps of:
This is determined by comparing the similarity detected in the first similarity detection step with a predetermined threshold value.

In the image processing method for pattern matching according to the present invention, in the step of determining a candidate area,
A predetermined number is selected from the larger similarity detected in the first similarity detection step.

The image processing apparatus according to the present invention creates a reference DCT pattern by performing discrete cosine transform on the reference pattern, sets a comparison area of the same size as the reference pattern from the input DCT image, and uses DCT coefficients. The similarity between the reference DCT pattern and the comparison area is detected from the inner product, the normalized correlation, and the like, and it is determined whether or not all the reference patterns whose positions have been changed in the input image have been compared with the entire area having the same size area. Then, the similarity detected in the similarity detection processing is compared with a predetermined threshold value. If there is a similarity value greater than the threshold value, the similarity to the reference pattern is found in the input DCT image. When it is determined that a pattern is included and all the similarities are equal to or smaller than the threshold, the input DC
The T image is determined not to include a pattern similar to the reference pattern.

An image processing apparatus according to the present invention performs a discrete cosine transform on a reference pattern to create a reference DCT pattern, selects one from a plurality of input DCT images, and selects one of the selected input DCT images. , A comparison area of the same size as the reference pattern is set, the similarity between the reference DCT pattern and the comparison area is detected from the inner product using the DCT coefficient, the normalized correlation, and the like. It is determined whether or not the comparison has been made with all regions having the same size region as the reference pattern, and the similarity detected in the similarity detection processing is compared with a predetermined threshold value. If there is a degree, input DC
When it is determined that a pattern similar to the reference pattern is included in the T image, and all the similarities are equal to or less than the threshold,
It is determined that the input DCT image does not include a pattern similar to the reference pattern, it is determined whether all the input DCT images have been selected, and if not all the images have been selected, the input image selection process is performed. Return to another input DC
This is to reselect the T image.

An image processing apparatus according to the present invention performs a discrete cosine transform on a reference pattern to create a reference DCT pattern, selects a DCT coefficient from a low frequency range, and inputs
A comparison area of the same size as the reference pattern is set from the CT image, and the similarity between the reference DCT pattern and the comparison area is detected using only selected DCT coefficients and compared with all predetermined areas. If not, the process returns to the first comparison region setting process, and the position is shifted from the last time to set the comparison region again, and the candidate is determined based on the first similarity. A region is determined, a DCT coefficient is selected from a high-frequency region, a comparison region is selected from candidate regions, and a similarity between the reference DCT pattern and the comparison region is detected using only selected DCT coefficients. It is determined whether or not all the candidate areas have been compared, and if not compared with all the areas, the process returns to the second comparison area setting process, and a candidate area different from the previous area is selected again to detect Similarity and predetermined threshold The values are compared, and if a similarity greater than the threshold value is present, it is determined that a pattern similar to the reference pattern is included in the input image. The image is determined not to include a pattern similar to the reference pattern.

[0019]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described below. Embodiment 1 FIG. FIG. 1 is a block diagram showing an image processing apparatus according to Embodiment 1 of the present invention, and FIG. 2A is an explanatory diagram showing a reference pattern in an image processing method for pattern matching according to Embodiment 1 of the present invention. 2 (b) is an explanatory diagram showing an input image in the image processing method for pattern matching according to the first embodiment of the present invention, and FIG. 3 shows an image processing method for pattern matching according to the first embodiment of the present invention. It is a flowchart. FIG. 4 is a configuration diagram showing an input DCT image of the image processing method for pattern matching according to the first embodiment of the present invention.

In FIG. 1, reference numeral 1 denotes a DC
Reference DCT for applying T to create reference DCT pattern 13
Pattern creation means 2, reference DCT pattern creation means 1
Area setting means 3 for setting a comparison area 18 divided for each DCT block 14 with respect to an input DCT image 17 input from the CPU 3, and a comparison area 3 between the reference pattern 11 and the comparison area 18 set by the comparison area setting means 2. The similarity detecting means 4 for detecting the similarity using the DCT coefficient 4 repeatedly determines whether or not the input DCT image 17 has been compared with all the areas of the same size as all the reference patterns 11 whose positions have been changed. The determining means 5 is a similarity determining means for determining whether or not a pattern similar to the reference pattern 11 is included by comparing the similarity detected by the similarity detecting means 3 with a threshold value.

Next, the operation will be described. First, criterion D
In the CT pattern creating means 1, the reference pattern 11
CT is performed to create a reference DCT pattern 13 (step ST1). The size 12 (16) shown in FIG.
× 16) with respect to the reference pattern 11 in units of 8 × 8
When T is performed, a reference DCT pattern 13 including four DCT blocks 14 is created as shown in FIG. Each element of the DCT block 14 is called a DCT coefficient 16 and represents a spatial frequency component in a corresponding block of the original image. Note that when the reference DCT pattern 13 is created, the DCT
However, since the reference pattern 11 is generally much smaller than the input image, the amount of calculation does not matter. Next, in the comparison area setting means 2, the comparison area 1 having the same size as the reference pattern 11 is input from the input DCT image 17.
8 is set (step ST2). As shown in FIG. 4, the input DCT image 17 is divided into blocks of 8.times.8, and the comparison area 18 is set at the position of each block.

Next, in the similarity detecting means 3, the reference D
The similarity between the CT pattern 13 and the comparison area 18 is detected (step ST3). The similarity is obtained by an inner product using a DCT coefficient, a normalized correlation, or the like. Subsequently, the repetition determination means 4 determines whether or not the comparison has been made with all the predetermined regions (step ST4). All the predetermined regions are, for example, regions having the same size as all the reference patterns 11 whose positions have been changed in the input image. If the comparison has not been made with all the areas, the process returns to step ST2, and the position is shifted from the previous time, and the comparison area 18 is set again.

Finally, the similarity determination means 5 compares the similarity detected by the similarity detection means 3 with a predetermined threshold value. Reference pattern 1 in DCT image 17
It is determined that a pattern similar to 1 is included (step ST5). On the other hand, when all the similarities are equal to or smaller than the threshold, it is determined that the input DCT image 17 does not include a pattern similar to the reference pattern 11.

As described above, according to the first embodiment, since the matching processing is performed by directly using the DCT coefficient 16 of the input DCT image 17, the preprocessing of performing the inverse DCT on the entire input DCT image 17 is performed. unnecessary. For this reason,
The effect is obtained that the matching processing for the input DCT image 17 can be efficiently performed.

Embodiment 2 FIG. 5 is a block diagram showing an image processing apparatus according to Embodiment 2 of the present invention, and FIG. 6 is a flowchart showing an image processing method for pattern matching according to Embodiment 2 of the present invention. The same reference numerals as those in the first embodiment denote the same or corresponding parts, and a description thereof will not be repeated. 6 is an input image selecting means for selecting one from a plurality of input DCT images 17, and 7 is all input DCT images.
This is an end determination unit that determines whether the image 17 has been selected.

Next, the operation will be described. Embodiment 2
Is performed on a plurality of input DCT images 17
This is for determining whether or not a pattern similar to the reference pattern 11 is included. First, the reference DCT pattern creating means 1 applies DCT to the reference pattern 11 to create a plurality of reference DCT patterns 13 (step ST1).
In the input image selecting means 6, a plurality of input DCT images 1
One is selected from 7 (step ST6). Next, the comparison area setting means 2 sets a comparison area 18 having the same size as the reference pattern 11 from the input DCT image 17 selected by the input image selection means 6 (step ST).
2). Next, the similarity detection means 3 detects the similarity between the reference DCT pattern 13 and the comparison area 18 (step ST3). Subsequently, in the repetition determination means 4,
It is determined whether the comparison has been made with all the predetermined areas (step ST4). If the comparison has not been made with all the areas, the process returns to step ST2, and the position is shifted from the previous time to set the comparison area 18 again.

Subsequently, the similarity determination means 5 compares the detected similarity with a predetermined threshold value (step ST5). As a result, if a similarity greater than the threshold value exists, the selected input DCT image 1
It is determined that a pattern similar to the reference pattern 11 is included in 7. On the other hand, if all the similarities are equal to or smaller than the threshold, it is determined that the selected input DCT image 17 does not include a pattern similar to the reference pattern 11. Finally, the end determination means 7 determines whether all input DCT images 17 have been selected (step ST7).
If all images have not been selected, step ST
6, the input DCT image 17 different from the previous one is selected again.

As described above, according to the second embodiment, there is no need to input the creation of the reference DCT pattern 13 and execute it for each image.
For example, the amount of calculation in the matching process can be reduced.

Embodiment 3 FIG. 7 is a configuration diagram showing a DCT area used in the image processing device according to the third embodiment of the present invention. In the figure, the same reference numerals as those in the first embodiment denote the same or corresponding parts, and a description thereof will not be repeated. The configuration of the image processing apparatus according to the third embodiment of the present invention is the same as that of the first embodiment, and a description thereof will be omitted.

Next, the operation will be described. FIG. 8 is a flowchart showing an image processing method for pattern matching according to Embodiment 3 of the present invention. First, a DCT is applied to the reference pattern to create a reference DCT pattern 13. At this time, as shown in FIG. 7, while changing the shift (x-direction shift 32 and y-direction shift 33) between the reference point of the DCT area and the reference point of the reference pattern 11, a plurality of reference D
A CT pattern 13 is created (step ST1). Next, a comparison area 18 having the same size as the reference pattern 11 is set from the input DCT image 17 (step ST2). Next, the similarity between the plurality of reference DCT patterns 13 and the comparison area 18 is detected. Since there are a plurality of reference DCT patterns 13, a plurality of similarities are also obtained, and the maximum value is set as the similarity to the comparison area 18 (step ST3).

Subsequently, it is determined whether or not the comparison has been made with all the predetermined areas. If not compared with all areas, the process returns to step ST2, and the position is shifted from the previous time to set a comparison area again (step ST4). Finally, compare the detected similarity with a predetermined threshold,
As a result of the comparison, when there is a similarity greater than the threshold value, it is determined that a pattern similar to the reference pattern 11 is included in the input image. On the other hand, if all the similarities are equal to or smaller than the threshold value, it is determined that the input DCT image 17 does not include a pattern similar to the reference pattern 11 (step ST5). In general, the reference pattern 11
Are the same, the DCT image 31 is different if the position of the block division for performing the DCT is different.

As described above, according to the third embodiment, since a plurality of reference DCT patterns 13 having different block division positions are prepared, the matching processing independent of the block division position of the DCT image 31 is performed. And the like.

Embodiment 4 FIG. FIG. 9 is a block diagram showing an image processing apparatus according to Embodiment 4 of the present invention, and FIG. 10 is a flowchart showing an image processing method for pattern matching according to Embodiment 4 of the present invention. In the figure,
The same reference numerals as those in the first embodiment denote the same or corresponding parts, and a description thereof will not be repeated. Reference numeral 8 denotes DCT coefficient selection means for selecting a DCT coefficient used for similarity detection. For example, when the size of the DCT area 31 is 8 × 8, the number of DCT coefficients is 64. Select

Next, the operation will be described. First, DCT
The DCT coefficient used for the similarity detection is selected by the coefficient selecting means 8 (step ST8), and the reference DCT is applied to the reference pattern by the reference DCT pattern creating means 1 to obtain a necessary DCT coefficient in the reference DCT pattern 13. (Step ST1). Here, it is only necessary to find eight coefficients in the low frequency region. Next, a comparison area 18 having the same size as the reference pattern 11 is set from the input DCT image 17 in the comparison area setting means 2 (step ST2). Next, the similarity detection means 3 detects the similarity between the reference DCT pattern 13 and the comparison area 18 using only selected DCT coefficients (step ST3), and repeats the determination means 4.
In step ST4, it is determined whether or not comparison has been made with all the predetermined regions.

If the comparison has not been made with all the areas, the process returns to step ST2 and the position is shifted from the previous area to make the comparison area 1
Set 8 again. Finally, in the similarity determination means 5,
The detected similarity is compared with a predetermined threshold. As a result, when there is a similarity greater than the threshold, it is determined that the input image includes a pattern similar to the reference pattern. On the other hand, if all the similarities are equal to or smaller than the threshold value, it is determined that the input image does not include a pattern similar to the reference pattern 11 (step S).
T5). In this image processing method, for example, when eight coefficients are selected, the amount of calculation for obtaining the similarity is all D
There is an advantage that it becomes 1/8 as compared with the case where the CT coefficient is used.

As described above, according to the fourth embodiment, it is possible to reduce the amount of calculation for the matching process on the input DCT image 17 and the amount of calculation for creating the reference DCT pattern 13.

Embodiment 5 FIG. 11 is a block diagram showing an image processing apparatus according to Embodiment 5 of the present invention, and FIG. 12 is a flowchart showing an image processing method for pattern matching according to Embodiment 5 of the present invention. In the figure, the same reference numerals as those in the first and fourth embodiments denote the same or corresponding parts, and thus the description thereof will be omitted. Numeral 9 is a calculation accuracy setting means for setting the calculation accuracy in the similarity detection in accordance with the frequency of the selected DCT coefficient. This is because when DCT is performed on a general image, D
Although high precision is required for the expression of CT coefficients, D
This reflects the characteristic that low precision is sufficient for expressing the CT coefficient.

Next, the operation will be described. First, the DCT coefficient selection means 8 selects a DCT coefficient used for similarity detection (step ST8). Next, the calculation accuracy setting means 9 sets the calculation accuracy in the similarity detection according to the frequency of the selected DCT coefficient (step ST).
9). Next, in the reference DCT pattern creating means 1,
DCT is applied to the reference pattern to obtain the reference DCT pattern 13
The required DCT coefficient in is obtained (step ST1).
Next, the comparison area setting means 2 sets a comparison area of the same size as the reference pattern from the input DCT image (step ST2). Next, the similarity detection means 3 detects the similarity between the reference DCT pattern 13 and the comparison area 18 using only selected DCT coefficients (step ST3).

Subsequently, the repetition determination means 4 determines whether or not the comparison has been made with all the predetermined areas (step S).
T4) If the comparison has not been performed with all the regions, the process returns to step ST2 and the position is shifted from the previous region to compare the region 18
Set again. Finally, the similarity determination means 5 compares the detected similarity with a predetermined threshold. As a result, when there is a similarity greater than the threshold, it is determined that the input image includes a pattern similar to the reference pattern. On the other hand, if all the similarities are equal to or smaller than the threshold value, it is determined that the input image does not include a pattern similar to the reference pattern (step ST
5).

As described above, according to the fifth embodiment, the power consumption of execution by H / W can be reduced by appropriately reducing the calculation accuracy in the matching processing, and the matching performance is not reduced. There are obtained effects such as reduction in power consumption of execution by H / W.

Embodiment 6 FIG. FIG. 13 is a block diagram showing an image processing apparatus according to Embodiment 6 of the present invention, and FIG. 14 is a flowchart showing an image processing method for pattern matching according to Embodiment 6 of the present invention. In the figure, the same reference numerals as those in the first, fourth, and fifth embodiments denote the same or corresponding parts, and a description thereof will not be repeated. Reference numeral 20 denotes a first DCT coefficient selection unit that selects a DCT coefficient used in the first similarity detection unit 22 from a low frequency band. Reference numeral 21 denotes a second setting unit that sets a comparison area 18 having the same size as the reference pattern 11 from the input DCT image 17. The first comparison area setting means 22 is a first similarity detection means for detecting the similarity between the reference DCT pattern 13 and the comparison area 18 using only selected DCT coefficients. This is a first repetition determination means for determining whether or not the comparison has been made.

Reference numeral 24 denotes a candidate area determining means for determining a candidate area based on the first similarity, and reference numeral 25 denotes a second D for selecting a DC coefficient used in the second similarity detecting means 27 from a high frequency band.
CT coefficient selecting means 26 selects comparison region 1 from the candidate regions.
The second comparison area setting means for selecting 8 is a reference DC.
Second similarity detecting means for detecting the similarity between the T pattern 13 and the comparison area 18 using only selected DCT coefficients, and a second similarity detection means 28 for determining whether or not all the candidate areas have been compared Is a repetition determination means.

Next, the operation will be described. First, the reference DC
In the T pattern creating means 1, D
CT is performed to create a reference DCT pattern 13 (step ST1), and the first DCT coefficient selection means 20 selects a DCT coefficient used by the first similarity detection means 22 from a low frequency band (step ST10). Next, in the first comparison area setting means 21, a comparison area 18 having the same size as the reference pattern 11 is set from the input DCT image 17 (step ST11).
In 2, the similarity between the reference DCT pattern 13 and the comparison area 18 is detected using only selected DCT coefficients (step ST12). Subsequently, the first repetition determination means 23 determines whether or not the comparison has been made with all the predetermined regions (step ST13). If the comparison has not been made with all the regions, the process returns to step ST11 and returns to step ST11. Is shifted to set the comparison area 18 again.

Subsequently, in the candidate area determining means 24,
A candidate area is determined based on the first similarity (step S
T14). As a determination method, there are a method of selecting a region having a similarity equal to or higher than a predetermined threshold value, and a method of selecting a predetermined number of regions from the one with the highest similarity. Next, the second DCT coefficient selection unit 25 selects a DCT coefficient used in the second similarity detection unit 27 from a high frequency band (step ST15). Next, the comparison area 18 is selected from the candidate areas by the second comparison area setting means 26 (step ST16). Next, the second similarity detection means 27 detects the similarity between the reference DCT pattern 13 and the comparison area 18 using only selected DCT coefficients (step ST17).

Subsequently, the second repetition determination means 28 determines whether or not all the candidate areas have been compared (step ST18). If not compared with all the regions, the process returns to step ST16, and a candidate region different from the previous region is selected again. Finally, the similarity determination means 5 compares the detected similarity with a predetermined threshold. As a result, when there is a similarity greater than the threshold, it is determined that the input image includes a pattern similar to the reference pattern. On the other hand, if all the similarities are equal to or smaller than the threshold value, it is determined that the input image does not include a pattern similar to the reference pattern (step ST
5). The low frequency range of the DCT coefficients reflects the coarse features of the original image. On the other hand, the high frequency range reflects the fine features of the original image. Therefore, in this image processing method, a low-frequency DCT coefficient is used for similarity detection at the stage of narrowing down candidates, and a high-frequency DCT coefficient is used for similarity detection for final determination. .

As described above, according to the sixth embodiment, a matching process for a DCT image having a large size can be realized with a small amount of calculation. If it is found that there is no similar pattern in the stage of narrowing down candidates, the second stage is executed. It is possible to obtain an effect that it can be omitted, and it can be applied to an application in which setting of a threshold for narrowing down candidates is difficult.

[0047]

As described above, according to the present invention, a reference DCT pattern is created by performing discrete cosine transform on a reference pattern, and a comparison area having the same size as the reference pattern is set from an input DCT image. Dot product using coefficients,
The similarity between the reference DCT pattern and the comparison area is detected from the normalized correlation or the like, and it is determined whether or not all the reference patterns whose positions have been changed in the input image have been compared with all areas having the same size area, The detected similarity is compared with a predetermined threshold, and if a similarity greater than the threshold is present, it is determined that a pattern similar to the reference pattern is included in the input DCT image, When all the similarities are equal to or smaller than the threshold value, it is determined that the input DCT image does not include a pattern similar to the reference pattern. Therefore, it is possible to efficiently perform the matching process on the DCT image. There is an effect that can be done.

According to the present invention, a reference DCT pattern is created by performing discrete cosine transform on the reference pattern, one of the plurality of input DCT images is selected, and one of the input images selected in the input image selecting step is selected. A comparison area of the same size as the reference pattern is set from the DCT image,
The similarity between the reference DCT pattern and the comparison area is detected from the inner product using the coefficients, the normalized correlation, and the like, and the similarity is calculated using only some selected DCT coefficients whose positions have been changed in the input image. It is determined whether or not comparison has been performed with all regions having the same size region as the reference pattern to be detected, and the detected similarity is compared with a predetermined threshold value. Input DC, if present
When it is determined that a pattern similar to the reference pattern is included in the T image, and all the similarities are equal to or less than the threshold,
It is determined that the input DCT image does not include a pattern similar to the reference pattern, it is determined whether or not all input DCT images have been selected, and if not all images have been selected, the process proceeds to the input image selection step. Return Input D different from last time
Since the system is configured so that the CT image is selected again,
There is an effect that the amount of calculation of the matching processing for the T image can be reduced.

According to the present invention, in the step of creating the reference DCT pattern, the plurality of reference DCTs are shifted while changing the shifts in the X-direction shift and the Y-direction shift of the reference point of the area where discrete cosine transform is performed and the reference point of the reference pattern. In the step of creating a pattern and detecting the degree of similarity, the method is configured to detect the degree of similarity between the plurality of reference DCT patterns and the comparison area, so that it is possible to realize a matching process that does not depend on the block division position of the DCT image. There is an effect that can be done.

According to the present invention, D used for similarity detection
The method includes a coefficient selection step of selecting a CT coefficient, and the similarity detection step is configured to detect the similarity using only some of the selected DCT coefficients. Has the effect of being able to reduce

According to the present invention, between the step of selecting the coefficient and the step of creating the reference DCT pattern, the calculation accuracy in the similarity detection is set according to the selected DCT coefficient. There is an effect that the power consumption of execution by H / W can be reduced by appropriately reducing the calculation accuracy in.

According to the present invention, in the calculation accuracy setting step, the calculation accuracy in the similarity detection is set in accordance with the frequency of the selected DCT coefficient. There is an effect that one power consumption of execution by W can be reduced.

According to the present invention, in the step of creating the reference DCT pattern, discrete cosine transform is performed on the reference pattern to create only DCT coefficients necessary for similarity detection. This has the effect of reducing the amount of calculation of.

According to the present invention, a discrete DCT pattern is created by performing discrete cosine transform on the reference pattern,
The DCT coefficient is selected from the low frequency range, and a comparison area having the same size as the reference pattern is set from the input DCT image.
When the similarity between the CT pattern and the comparison area is detected using only some of the selected DCT coefficients, and it is determined whether or not the similarity is compared with all the predetermined areas, and the comparison is not made with all the areas. Returns to the first comparison area setting step, resets the comparison area by shifting the position from the previous time, determines the candidate area based on the first similarity, selects the DCT coefficient from the high frequency area, Select the comparison area from
Some DCs whose similarity between the pattern and the comparison area is selected
Detection is performed using only the T coefficient, and it is determined whether or not the comparison has been made with all of the candidate regions.
Returning to the second comparison area setting step, a candidate area different from the previous area is selected again, the detected similarity is compared with a predetermined threshold, and if there is a similarity larger than the threshold, It is determined that a pattern similar to the reference pattern is included in the input image, and if all the similarities are equal to or less than the threshold value, it is determined that the input image does not include a pattern similar to the reference pattern. Thus, there is an effect that the matching processing for a DCT image having a large size can be realized with a small amount of calculation.

According to the present invention, in the candidate area determination step, the similarity detected in the first similarity detection step is determined by comparing the similarity detected with the predetermined threshold value. If it is found that there is no similar pattern at the stage, there is an effect that the second stage can be omitted.

According to the present invention, in the step of determining a candidate area, a predetermined number is selected from the largest similarity detected in the first similarity detection step. There is an effect that the setting of the threshold value can be applied to an application that is difficult.

According to the present invention, a reference DCT pattern is created by performing discrete cosine transform on the reference pattern, a comparison area of the same size as the reference pattern is set from the input DCT image, and the inner product and the normal product using the DCT coefficient are set. The similarity between the reference DCT pattern and the comparison area from
It is determined whether or not all the reference patterns whose positions have been changed in the input image have been compared with all regions having the same size region, and the similarity detected in the similarity detection processing and a predetermined threshold value are determined. Are compared, and if a similarity greater than this threshold exists, it is determined that a pattern similar to the reference pattern is included in the input DCT image, and if all the similarities are equal to or smaller than the threshold, Since the input DCT image is configured to be determined not to include a pattern similar to the reference pattern, there is an effect that the matching processing on the DCT image can be efficiently performed.

According to the present invention, the reference DCT pattern is created by performing discrete cosine transform on the reference pattern, one of the plurality of input DCT images is selected, and the reference pattern is selected from the selected one input DCT image. A comparison area having the same size as the above is set, and the similarity between the reference DCT pattern and the comparison area is detected from the inner product using the DCT coefficient, the normalized correlation, and the like. It is determined whether or not all regions having the same size region have been compared, and the similarity detected in the similarity detection process is compared with a predetermined threshold value. If it is determined that the input DCT image includes a pattern similar to the reference pattern,
If all the similarities are equal to or less than the threshold value, it is determined that the input DCT image does not include a pattern similar to the reference pattern, and it is determined whether or not all the input DCT images have been selected. If is not selected, the process returns to the input image selection process and the input DCT image different from the previous one is selected again, so that the amount of calculation in the matching process for a plurality of DCT images can be reduced. There is.

According to the present invention, a reference DCT pattern is created by performing discrete cosine transform on the reference pattern,
The DCT coefficient is selected from the low frequency range, and a comparison area having the same size as the reference pattern is set from the input DCT image.
When the similarity between the CT pattern and the comparison area is detected using only some of the selected DCT coefficients, and it is determined whether or not the similarity is compared with all the predetermined areas, and the comparison is not made with all the areas. Returns to the first comparison area setting process, resets the comparison area by shifting the position from the previous time, determines the candidate area based on the first similarity, selects the DCT coefficient from the high frequency area, Is selected from among the DCT patterns, and the similarity between the reference DCT pattern and the comparison area is selected from the selected DCT patterns.
Detection is performed using only the coefficient, and it is determined whether or not the comparison has been made with all of the candidate regions. If the comparison has not been made with all of the regions, the process returns to the second comparison region setting process to select a candidate region different from the previous region. Reselect and compare the detected similarity with a predetermined threshold. If there is a similarity greater than the threshold, determine that the input image contains a pattern similar to the reference pattern. If all the similarities are equal to or smaller than the threshold value, it is determined that the input image does not include a pattern similar to the reference pattern. There is an effect that can be realized with.

[Brief description of the drawings]

FIG. 1 is a configuration diagram illustrating an image processing apparatus according to a first embodiment of the present invention;

FIG. 2A is an explanatory view showing a reference pattern in an image processing method for pattern matching according to the first embodiment of the present invention, and FIG. 2B is an image for pattern matching according to the first embodiment of the present invention; FIG. 4 is an explanatory diagram illustrating an input image in a processing method.

FIG. 3 is a flowchart showing an image processing method for pattern matching according to the first embodiment of the present invention.

FIG. 4 is a configuration diagram showing an input DCT image of the image processing method for pattern matching according to the first embodiment of the present invention;

FIG. 5 is a configuration diagram illustrating an image processing apparatus according to a second embodiment of the present invention.

FIG. 6 is a flowchart illustrating an image processing method for pattern matching according to Embodiment 2 of the present invention;

FIG. 7 is a configuration diagram showing a DCT area used in an image processing device according to a third embodiment of the present invention.

FIG. 8 is a flowchart showing an image processing method for pattern matching according to Embodiment 3 of the present invention.

FIG. 9 is a configuration diagram illustrating an image processing apparatus according to a fourth embodiment of the present invention.

FIG. 10 is a flowchart showing an image processing method for pattern matching according to Embodiment 4 of the present invention.

FIG. 11 is a configuration diagram illustrating an image processing apparatus according to a fifth embodiment of the present invention.

FIG. 12 is a flowchart showing an image processing method for pattern matching according to Embodiment 5 of the present invention.

FIG. 13 is a configuration diagram illustrating an image processing apparatus according to a sixth embodiment of the present invention.

FIG. 14 is a flowchart showing an image processing method for pattern matching according to Embodiment 6 of the present invention.

FIG. 15 is a flowchart showing a conventional image processing method for pattern matching shown in Pattern Understanding (Yoshiaki Shirai, Ohmsha).

FIG. 16 is an explanatory diagram showing a reference pattern in a conventional image processing method for pattern matching.

FIG. 17 is an explanatory diagram showing an input image in a conventional image processing method for pattern matching.

[Explanation of symbols]

1 reference DCT pattern creation means, 2 comparison area setting means, 3 similarity detection means, 4 repetition determination means, 5
Similarity determination means, 6 input image selection means, 7 end determination means, 8 DCT coefficient selection means, 11 reference patterns, 1
3 reference DCT pattern, 17 input DCT images, 18
Comparison area, 20 first DCT coefficient selecting means, 21
First comparison area setting means, 22 first similarity detection means, 23 first repetition determination means, 24 candidate area determination means, 25 second DCT coefficient selection means, 26 second comparison area setting means, 27 Second similarity detecting means, 28
Second repetition determination means, 31 DCT images.

Claims (13)

[Claims] 1. A reference DCT pattern creating step of creating a reference DCT pattern by applying a discrete cosine transform to a reference pattern; a comparison area setting step of setting a comparison area of the same size as the reference pattern from an input DCT image; A similarity detection step of detecting a similarity between the reference DCT pattern and the comparison area from an inner product using coefficients, a normalized correlation, and the like; an area having the same size as all the reference patterns whose positions have been changed in the input image. A repetition determining step of determining whether or not the similarity is compared with the entire area having the similarity; and comparing the similarity detected in the similarity detecting step with a predetermined threshold value; If there is a pattern, it is determined that a pattern similar to the reference pattern is included in the input DCT image, If the similarity is below the threshold, an image processing method for pattern matching in the input DCT image has a determining similarity determination step and does not include the pattern similar to the reference pattern. 2. A reference DCT pattern creating step of creating a reference DCT pattern by applying a discrete cosine transform to the reference pattern; an input image selecting step of selecting one from a plurality of input DCT images; A comparison area setting step of setting a comparison area of the same size as the reference pattern from one of the input DCT images selected by the above-mentioned method; and an inner product using DCT coefficients, a normalized correlation, and the like. A similarity detection step of detecting a similarity; a repetition determination step of determining whether or not all the reference patterns whose positions have been changed in the input image are compared with an entire area having the same size area; The similarity detected in the detection step is compared with a predetermined threshold, and the If there is a similar degree of similarity, it is determined that a pattern similar to the reference pattern is included in the input DCT image, and if all the similarities are equal to or less than the threshold, the input DCT image A similarity determination step of determining that a pattern similar to the reference pattern is not included; determining whether all of the input DCT images have been selected; if not all images have been selected, An image processing method for pattern matching, comprising: returning to the input image selecting step and re-selecting the input DCT image different from the previous one. 3. In the reference DCT pattern creating step,
A plurality of reference DCT patterns are created while changing the shift of the X-direction shift and the Y-direction shift between the reference point of the area where the discrete cosine transform is performed and the reference point of the reference pattern. 2. The image processing method for pattern matching according to claim 1, wherein a similarity with the comparison area is detected. 4. A method according to claim 1, further comprising a coefficient selection step of selecting a DCT coefficient used for similarity detection, wherein the similarity detection step detects the similarity using only some selected DCT coefficients. Claim 1
An image processing method for the described pattern matching. 5. The method according to claim 4, further comprising, between the coefficient selecting step and the reference DCT pattern creating step, an operation accuracy setting step of setting the operation accuracy in the similarity detection in accordance with the selected DCT coefficient. An image processing method for the described pattern matching. 6. The image processing method for pattern matching according to claim 5, wherein in the calculation accuracy setting step, the calculation accuracy in the similarity detection is set according to the frequency of the selected DCT coefficient. 7. The image processing for pattern matching according to claim 4, wherein in the reference DCT pattern creating step, discrete cosine transform is performed on the reference pattern to create only DCT coefficients necessary for similarity detection. Method. 8. A reference DCT pattern creating step of applying a discrete cosine transform to the reference pattern to create a reference DCT pattern; a first DCT coefficient selecting step of selecting a DCT coefficient from a low frequency band; A first comparison area setting step of setting a comparison area having the same size as the reference pattern; and a first detection step of detecting a similarity between the reference DCT pattern and the comparison area using only selected DCT coefficients. The similarity detection step determines whether or not the comparison has been made with all of the predetermined areas. If the comparison has not been made with all of the areas, the procedure returns to the first comparison area setting step, and the position is shifted from the previous time to shift the position. A first iterative determination step of resetting the comparison area; a candidate area determination step of determining a candidate area based on the first similarity; A second DCT coefficient selection step of selecting from the following: a second comparison area setting step of selecting the comparison area from the candidate areas; and a part in which the similarity between the reference DCT pattern and the comparison area is selected. A second similarity detection step of detecting using only the DCT coefficients of the above, and determining whether or not all of the candidate areas have been compared, and if not comparing with all of the areas, setting the second comparison area A second repetition determination step of returning to the step and selecting a candidate area different from the previous one; and comparing the detected similarity with a predetermined threshold value. It is determined that a pattern similar to the reference pattern is included in the input image, and if all the similarities are equal to or less than the threshold value, the input image includes a pattern similar to the reference pattern. An image processing method for pattern matching, comprising a similarity determination step of determining that the pattern is not rare. 9. The pattern matching method according to claim 8, wherein the candidate area determination step is performed by comparing the similarity detected in the first similarity detection step with a predetermined threshold value. Image processing method. 10. The image for pattern matching according to claim 8, wherein in the candidate area determination step, a predetermined number is selected from the larger similarities detected in the first similarity detection step. Processing method. 11. A reference DCT pattern creating means for applying a discrete cosine transform to a reference pattern to create a reference DCT pattern; a comparison area setting means for setting a comparison area of the same size as the reference pattern from an input DCT image; A similarity detecting means for detecting a similarity between the reference DCT pattern and the comparison area from an inner product using a coefficient, a normalized correlation, and the like; and an area having the same size as all the reference patterns whose positions have been changed in the input image. And a repetition determining means for determining whether or not the comparison has been made with the entire area having the similarity. The similarity detected by the similarity detecting means is compared with a predetermined threshold value. If there is a pattern, it is determined that a pattern similar to the reference pattern is included in the input DCT image, and all the similarities are higher. In the case of less than or equal to the threshold value,
An image processing apparatus comprising: a similarity determination unit configured to determine that the input DCT image does not include a pattern similar to the reference pattern. 12. A reference DCT pattern creating means for applying a discrete cosine transform to a reference pattern to create a reference DCT pattern; an input image selecting means for selecting one from a plurality of input DCT images; Comparison area setting means for setting a comparison area of the same size as the reference pattern from one of the input DCT images selected by the above-mentioned method, and a comparison between the reference DCT pattern and the comparison area based on inner products using DCT coefficients, normalized correlation, and the like. Similarity detection means for detecting similarity; repetition determination means for determining whether or not all the reference patterns whose positions have been changed in the input image are compared with all areas having the same size area; Comparing the above similarity detected by the detecting means with a predetermined threshold value, and when there is a similarity greater than this threshold value; Was determined to contain a similar pattern to the reference pattern in the input DCT image, if all of the similarity is less than the threshold value,
A similarity determination unit that determines that the input DCT image does not include a pattern similar to the reference pattern; and determines whether all of the input DCT images have been selected, and all images have not been selected. In this case, the image processing apparatus further includes an end determination unit that returns to the input image selection unit and selects another input DCT image different from the previous one. 13. A reference DCT pattern creating means for applying a discrete cosine transform to a reference pattern to create a reference DCT pattern; a first DCT coefficient selecting means for selecting a DCT coefficient from a low frequency band; First comparison area setting means for setting a comparison area having the same size as the reference pattern; and first detection means for detecting the similarity between the reference DCT pattern and the comparison area using only selected DCT coefficients. The similarity detection means determines whether or not the comparison has been made with all of the predetermined areas. First repetition determination means for resetting the comparison area, candidate area determination means for determining a candidate area based on the first similarity, and second DCT section for selecting a DCT coefficient from a high frequency range. Number selection means, second comparison area setting means for selecting the comparison area from the candidate areas, and similarity between the reference DCT pattern and the comparison area using only selected DCT coefficients. The second similarity detecting means to be detected determines whether or not all the candidate areas have been compared. If not, the processing returns to the second comparison area setting means and is different from the last time. Comparing the detected similarity with a predetermined threshold, and if there is a similarity greater than the threshold, the input image includes A similarity determination unit that determines that a pattern similar to the reference pattern is included, and when all the similarities are equal to or less than the threshold, determines that the input image does not include a pattern similar to the reference pattern. Images with Processing equipment.