JP3276813B2 - Search area limitation method in pattern matching - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a search area limiting method in pattern matching for searching for a matching position by moving a reference image with respect to a target image and calculating a degree of matching between the two images.

[0002]

2. Description of the Related Art In pattern matching, an operation of searching for the best matching position is called a search. In this search, there are various routes in the processing until reaching the best matching position. The most basic form of search is a method of performing search processing on all possible routes. For example, this corresponds to the case where the matching degree is calculated at all positions within the search range and the best matching position is obtained. In this method, when the search area is large, the amount of calculation becomes enormous and the processing time becomes long.

[0003] In order to perform the search efficiently, a method of progressing the search in a direction in which the evaluation function is improved (hill climbing method) has been developed. Also in the case of performing such a search, in general, the position having the highest matching degree from a plurality of positions evenly selected from the entire region of the search range is often set as the search start position of the hill-climbing method. Therefore, when the search area is large, the amount of calculation increases and the processing time increases.

[0004]

SUMMARY OF THE INVENTION It is an object of the present invention to provide a method for limiting a search area in pattern matching, in which a search range can be narrowed before searching for pattern matching and processing time can be reduced. And

[0005]

According to a first method of limiting a search area in pattern matching according to the present invention, a matching degree is calculated by moving a reference image with respect to a target image and calculating a degree of matching between the two images. In a search area limiting method in pattern matching for searching a position, a target image is divided into a plurality of blocks, an orthogonal transform is performed on each block, and a predetermined frequency is determined based on a transform coefficient obtained for each block. The method includes a step of extracting a block including the above high frequency component, wherein the extracted block is set as a significant block for pattern matching, and a search range of pattern matching is narrowed.

In a second search area limiting method in pattern matching according to the present invention, a matching pattern is searched for by moving a reference image with respect to a target image and calculating a matching degree between the two images. Dividing the reference image into a plurality of blocks and performing orthogonal transform on each block, based on the transform coefficients obtained for each block, the frequency components included in all the blocks. Of which
Extracting the frequency of the highest frequency component, determining a significant block determination frequency based on the extracted frequency, dividing the target image into a plurality of blocks, and performing orthogonal transformation on each block; And a step of extracting, from each block, a block containing a frequency component equal to or higher than a significant block determination frequency based on the transform coefficient obtained for each block. As a significant block, a search range for pattern matching is narrowed.

[0007]

Embodiments of the present invention will be described below with reference to the drawings.

[0008] Hereinafter, by moving the template image 2 shown in FIG. 3 with respect to the target image 1 shown in FIG. 6 and calculating the degree of matching between the two images, the search area is limited when searching for a matching position. The method will be described.

FIG. 1 shows a procedure for calculating a significant block determination value performed when a template image is determined.

First, as shown in FIG. 4, the template image 2 is divided into blocks of N × N pixels, for example, blocks of 8 × 8 pixels (step 1).

Next, the image of each block is decomposed into frequency components by a discrete cosine transform (DCT) (step 2). That is, as shown in FIG. 5, the image of each block is converted from the spatial coordinates f (x, y) to the frequency coordinates F (u, u).
v). Here, u indicates a horizontal spatial frequency, and v indicates a vertical spatial frequency.

In general, when a correlated image is decomposed into frequency components by discrete cosine transform, the low frequency components have a large value, but the high frequency components are almost zero. In other words, the higher the correlation image is, the more the frequency distribution after the conversion is shifted to the lower frequency side.

Therefore, among the blocks of the template image 2, the converted frequency distribution of the characteristic image block contains higher frequency components than the converted frequency distribution of the non-characteristic image block. In.

Next, based on the transform coefficients for all the blocks, the frequency (ua, va) of the highest frequency component among the frequency components having a value (transform coefficient) larger than 0 is obtained.
Is extracted (step 3).

Next, a distance La from the origin O of the frequency coordinates F (u, v) to the extracted highest frequency (ua, va) is calculated based on the following equation (step 4).

[0016]

## EQU1 ## La = (ua ² + va ² ) ^1/2

Next, a significant block discrimination value Lo (= k · La) is obtained by multiplying the obtained distance La by a predetermined constant k (step 5). As the constant k, a value in the range of 0 <k <1 is used. Here, 0.8 is set as k. The significant block discrimination value Lo thus obtained is stored.

FIG. 2 shows a search area limitation processing procedure performed before pattern matching is performed.

First, as shown in FIG.
It is divided into blocks of the same size as the blocks of the template image 2 (step 11).

Next, the image of each block is decomposed into frequency components by discrete cosine transform (DCT) (step 12). That is, the image of each block is represented by the spatial coordinates f
(X, y) is converted to frequency coordinates F (u, v).

Next, based on the transform coefficients for each block, the highest frequency (ub, vb) among the frequency components having a value (transform coefficient) larger than 0 in each block is extracted for each block. (Step 1
3).

Next, the frequency coordinate F
The distance Lb from the origin O of (u, v) to the extracted highest frequency (ub, vb) is obtained based on the following equation 2 (step 14).

[0023]

Lb = (ub ² + vb ² ) ^1/2

Next, for each block, it is determined whether or not the obtained distance Lb is greater than or equal to a significant block determination value Lo, and a block whose distance Lb is greater than or equal to the significant block determination value Lo is extracted as a significant block ( Step 15). FIG.
, The significant blocks are shown with diagonal lines.

When a significant block is obtained in this way, a pattern matching search is performed using a predetermined area including the significant block and its vicinity as a search area. As this search method, any search method may be used.

When the hill-climbing method is used as a search method, a search start point can be set from a significant block pattern to a block (indicated by reference numeral 10 in FIG. 8) corresponding to a corner of the significant block pattern. Good. In this case, the search time becomes very short.

Although the discrete cosine transform is used as the orthogonal transform in the above embodiment, other orthogonal transforms such as Walsh-Hadamard transform (WHT) and discrete Fourier transform (DFT) may be used.

[0028]

According to the present invention, the search range can be narrowed before searching for pattern matching, and the processing time can be shortened.

[Brief description of the drawings]

FIG. 1 is a flowchart illustrating a procedure of calculating a significant block determination value performed when a template image is determined.

FIG. 2 is a flowchart illustrating a search area limitation processing procedure performed before pattern matching is performed.

FIG. 3 is a schematic diagram illustrating an example of a template image.

FIG. 4 is a schematic diagram showing a state where a template image is divided into a plurality of blocks.

FIG. 5 shows an image of a block obtained by a discrete cosine transform.
FIG. 4 is a schematic diagram showing that space coordinates are converted into frequency coordinates.

FIG. 6 is a schematic diagram illustrating an example of a target image.

FIG. 7 is a schematic diagram showing a state in which a target image is divided into a plurality of blocks.

FIG. 8 is a schematic diagram illustrating a significant block among blocks of a target image.

[Explanation of symbols]

 1 Target image 2 Template image

────────────────────────────────────────────────── ─── Continuing on the front page (72) Inventor Takeshi Mogi 2-5-5 Keihanhondori, Moriguchi-shi, Osaka Inside Sanyo Electric Co., Ltd. (72) Inventor Hiroshi Kawada 2-5-2 Keihanhondori, Moriguchi-shi, Osaka No. 5 Sanyo Electric Co., Ltd. (56) References JP-A-4-195689 (JP, A) (58) Fields investigated (Int. Cl. ⁷ , DB name) G06T ⁷ /00-7/60 G06T 1 / 00

Claims (3)

(57) [Claims] 1. A search area limiting method in pattern matching for searching a matching position by moving a reference image with respect to a target image and calculating a matching degree between the two images, the method comprising: And performing an orthogonal transform on each block; and extracting a block containing a high-frequency component equal to or higher than a predetermined frequency based on a transform coefficient obtained for each block. A search area limiting method in pattern matching, characterized in that a search range of pattern matching is narrowed down by setting the selected block as a significant block of pattern matching. 2. A search area limiting method for pattern matching for searching a matching position by moving a reference image with respect to a target image and calculating a matching degree between the two images, wherein the reference image is divided into a plurality of blocks. And performing an orthogonal transform on each block; and extracting the highest frequency component frequency among the frequency components included in all the blocks based on the transform coefficient obtained for each block. Determining a significant block determination frequency based on the extracted frequency, dividing the target image into a plurality of blocks, performing an orthogonal transform on each block, and transform coefficients obtained for each block. From the blocks, the block containing the frequency component equal to or higher than the significant block determination frequency is Comprising the step of leaving, the extracted block as significant block of the pattern matching, the search area limited way in the pattern matching, wherein a search range of the pattern matching is narrowed. 3. The search area limiting method in pattern matching according to claim 1, wherein the orthogonal transform is a discrete cosine transform.