JPH1185996A - Separation degree filtering processor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To lower the calculation cost for a separation degree filtering process by calculating the sum of partial areas and the total sum of squares of the partial areas in an area in an image and calculating a separation degree from the current total sum and the current sum of the squares. SOLUTION: An initializing process part 11 calculates the total sum or the sum of squares in an area 1, an area 2, and the entire area through an initializing process at a first point of separation degree filtering. This calculation result is an initial value for a recurrence formula. Then a partial area total sum and square sum calculating process part 12 calculates the total sum and the sum of squares through a partial area sum and square sum calculating process for partial areas generated from the relation between the areas 1 and 2. Then a separation degree calculating process part 13 calculates a separation degree by using the total sum and square sum in the initializing process or partial areas. Then a control part 14 adds a temporary variable to decide whether or not the process ends. Consequently, the calculation cost is reduced to make the scale of a processing circuit small.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for speeding up a degree of separation filtering process, which is one of feature point extraction methods and contour detection methods used in face recognition and image monitoring using images.

[0002]

2. Description of the Related Art As a feature point extraction method using an image, a method using separation filtering has been proposed. This method is a method of extracting feature points such as a contour, a pupil, a nostril, and a mouth end with high accuracy from a difference in statistical properties between two regions in an image. For details of the separation filtering, refer to Japanese Patent Application No. 9-3.
This is described in 3704. The filtering of the degree of separation will be described later in this specification.

[0003]

SUMMARY OF THE INVENTION However, Japanese Patent Application No. Hei 9-3
Although the separation degree filtering algorithm described in Japanese Patent No. 3704 has generality, in the case of large-area processing such as full-screen processing, the high calculation cost becomes a problem.

[0004] It is an object of the present invention to reduce the operation cost of the separation degree filtering process and to solve the above-mentioned problems.

[0005]

According to a first aspect of the present invention, a partial area sum and a partial area square sum are calculated from the areas 1 and 2 in an image and separated from the partial area sum and the partial area square sum. In the separation degree filtering processing apparatus for extracting a feature point in the image by further calculating the degree, a partial area 1 which is an overlapping part of the current area 1 and the previous time point 1 which is a time point before the current time. Is obtained, and a partial region 2 which is an overlapping portion between the current region 2 and the previous time region 2 is obtained. The partial region 2 is a portion where the sum region generated from the current region 1 and the current time region 2 overlaps with the previous time sum region. Partial area processing means for obtaining a partial sum area, partial area 1 and partial area 2
And a partial area sum calculating means for calculating a partial area sum which is a sum of pixel values of the partial sum areas, and a partial area sum of squares which is a sum of squares of the pixel values of the partial areas 1, 2 and the partial area. From the sum of the partial area squares to be calculated, the sum of the pixel values of the area 1, the area 2 and the sum area at the previous time, and the sum of the partial areas, the current area 1 and the area 2 are calculated.
Current sum calculating means for obtaining a current sum that is a sum of pixel values of the sum area and a sum of squares of pixel values of the area 1, the area 2 and the sum area at the previous time, and a partial area Current square sum calculating means for calculating the current square sum which is the sum of the squares of the pixel values of the current region 1, the region 2 and the sum region from the square sum, and separating the current sum and the current square sum from each other And a separation degree calculating means for calculating the degree of separation.

The sum or square sum of the pixel values in the regions 1, 2 and the sum region at the previous time is obtained by calculating the initial value when calculating the degree of separation, and then sequentially stacking the time. calculate. Further, the partial area sum and the partial area square calculated from the areas 1 and 2 in the image are the separation degree masks described later.

According to a second aspect of the present invention, the sum of partial areas and the sum of partial areas are calculated from the areas 1 and 2 in the image, and the degree of separation is further calculated from the sum of the partial areas and the sum of partial areas. A region 1 setting means for setting a frame-shaped large region obtained by superimposing a middle rectangular region on a large rectangular region as a region 1; Area 2 setting means for setting a frame-shaped small area obtained by superimposing a rectangular small area in the area as an area 2, area 1, area 2, area 1 and area 2
Sum calculation means for calculating the sum of the pixel values of the sum area of the large area, the middle area of the rectangle and the small area of the rectangle, and the square sum of the pixel values of the areas 1, 2 and the large area,
A square sum calculating means for calculating from the sum of squares of the rectangular medium area and the rectangular small area, and a separation degree calculating means for calculating the degree of separation from the sum of the pixel values of the area 1, the area 2, and the sum area and the square sum. A filtering apparatus for filtering a degree of segregation.

According to a third aspect of the present invention, the sum calculation means includes:
The calculation of the sum of the pixel values of each rectangular area is performed separately in the horizontal and vertical directions, and the square sum calculation means calculates the sum of the squares of the pixel values of each rectangular area in the horizontal direction and the horizontal direction. 3. The separation degree filtering apparatus according to claim 2, wherein the separation degree filtering processing is performed separately in vertical processing.

According to a fourth aspect of the present invention, a horizontal partial area 1 which is an overlapping portion between the current horizontal area 1 and the previous horizontal area 1 which is a time before the current time is obtained.
A horizontal partial area 2 which is an overlapping portion between the current horizontal area 2 and the previous horizontal area 2 is obtained, and the horizontal sum area generated from the current horizontal areas 1 and 2 and the previous time Horizontal partial area processing means for obtaining a horizontal partial sum area that is an overlapping part with the horizontal sum area of the following; a current vertical area 1 and a previous vertical area 1 which is a time point earlier than the current vertical area 1; Is obtained as a vertical partial area 1 which is an overlapping part of
And a vertical partial area 2 which is an overlapping part of the previous vertical area 2 and a vertical sum area generated from the current vertical area 1 and the vertical area 2 and a vertical sum area at the previous time. A vertical partial area processing means for obtaining a vertical partial sum area which is an overlapping part of: and the sum calculating means, for each of the horizontal direction and the vertical direction, the pixel values of the partial area 1, the partial area 2, and the partial sum area. A partial area sum calculating means for calculating a partial area sum that is a sum;
A sum at a previous time point which is a sum of pixel values of the area 2 and the sum area;
Current sum calculating means for obtaining a current sum that is the sum of the pixel values of the current area 1, the area 2 and the sum area from the partial area sum, wherein the square sum calculating means comprises a horizontal direction and a vertical direction. A partial area sum-of-squares calculation means for calculating a partial area sum-of-squares which is a sum of squares of the pixel values of the partial areas 1, 2 and the partial sum areas; From the previous sum of squares of the sum of the pixel values of the sum area and the sum of the squares of the partial area, the current sum of squares, which is the sum of the squares of the pixel values of the area 1, the area 2, and the sum area at the current time 4. The apparatus for filtering the degree of separation according to claim 3, further comprising: a current sum-of-squares calculating means for obtaining the sum of squares.

According to a fifth aspect of the present invention, the sum of partial areas and the sum of partial areas are calculated from the areas 1 and 2 in the image, and the degree of separation is further calculated from the sum of the partial areas and the sum of partial areas. In the separation degree filtering processing method for extracting the feature points in the image, a partial area 1 which is an overlapping part of the current area 1 and the previous time area 1 which is a time point earlier than the current time area is obtained. A partial area 2 which is an overlapping portion between the area 2 at the previous time and the area 2 at the previous time is obtained, and a partial sum area which is an overlapping part between the sum area generated from the current area 1 and the area 2 and the sum area at the previous time are obtained. A partial area processing step; a partial area sum calculating step of calculating a partial area sum that is the sum of pixel values of the partial areas 1, 2 and the pixels of the partial areas 1, 2 and 2. value A partial area square sum calculation step of calculating a partial area square sum which is a square sum; a previous time total which is a sum of pixel values of the area 1, the area 2 and the sum area at a previous time; And a current sum calculation step for obtaining a current sum that is the sum of the pixel values of the current region 1, the region 2 and the sum region, and a square sum of the pixel values of the region 1, the region 2 and the sum region at the previous time. The present time 2 which is the sum of the squares of the pixel values of the current area 1, the area 2 and the sum area from the sum of the previous time square and the sum of the partial areas square
A separation filtering processing method comprising: a sum of squares calculation step; and a separation degree calculation step of calculating a separation degree from the current sum and the current square sum.

According to a sixth aspect of the present invention, a sum of partial areas and a sum of partial areas are calculated from areas 1 and 2 in an image, and a degree of separation is further calculated from the sum of partial areas and the sum of partial areas. In the separation degree filtering processing method for extracting the feature points in the image, a region 1 setting step of setting a frame-shaped large region obtained by superimposing a rectangular middle region on a rectangular large region as a region 1; An area 2 setting step of setting a frame-shaped small area obtained by superimposing a rectangular small area in the area as an area 2, and an area 1, an area 2, and an area 1
Sum calculation step of calculating the sum of the pixel values of the sum area of the area and the area 2 from the sum of the large rectangular area, the middle area of the rectangle and the small area of the rectangle, and the square sum of the pixel values of the area 1, the area 2 and the sum area Sum-of-squares calculation step of calculating from the sum of squares of the large area, the middle area of the rectangle, and the small area of the rectangle, and separation degree calculation of calculating the degree of separation from the sum of the pixel values of the areas 1, 2 and the sum area and the sum of squares And a separation degree filtering processing method characterized by comprising:

According to a seventh aspect of the present invention, in the total sum calculating step, the calculation of the total sum of the pixel values of each rectangular area is performed separately in horizontal and vertical directions. 7. The separation filtering method according to claim 6, wherein the calculation of the sum of squares of the pixel values of the rectangular area is performed separately in horizontal and vertical directions.

According to the present invention, a horizontal partial area 1 which is an overlapping portion of the current horizontal area 1 and a previous horizontal area 1 which is a time before the current time is obtained.
A horizontal partial area 2 which is an overlapping portion between the current horizontal area 2 and the previous horizontal area 2 is obtained, and the horizontal sum area generated from the current horizontal areas 1 and 2 and the previous time , A horizontal partial area processing step of obtaining a horizontal partial sum area that is an overlapping part with the horizontal sum area, and the current vertical area 1 and the previous vertical area that is a time point earlier than the current vertical area 1 1 is obtained, a vertical partial area 2 which is an overlapping part of the current vertical area 2 and the previous vertical area 2 is obtained, and the current vertical area 1 is obtained.
And a vertical partial area processing step of obtaining a vertical partial sum area that is an overlapping portion of the vertical sum area generated from the vertical area 2 and the vertical sum area at the previous time. A partial area sum calculation step of calculating a partial area sum that is the sum of the pixel values of the partial area 1, the partial area 2 and the partial sum area for each of the horizontal direction and the vertical direction; A current sum calculation step of calculating a current sum that is the sum of the pixel values of the current region 1, the region 2, and the sum region from the previous time sum that is the sum of the pixel values of the sum region and the sum of the partial regions. Calculating the sum of squares of the partial area, which is the sum of the squares of the pixel values of the partial area 1, the partial area 2, and the partial sum area for each of the horizontal direction and the vertical direction. Calculation step and previous time From the previous time point sum of squares of the pixel values of the area 1, the area 2 and the sum area, and the square sum of the pixel values of the area 1, the area 2 and the sum area at the present time A current square sum calculation step of obtaining a current square sum as a sum.

According to a ninth aspect of the present invention, a partial area sum and a partial area square sum are calculated from the area 1 and the area 2 in the image, and the degree of separation is further calculated from the partial area sum and the partial area square sum. In the recording medium on which the program for realizing the degree of separation filtering for extracting the feature points in the image is recorded, a portion which is an overlapping portion between the current region 1 and the previous region 1 which is a time point before the current time. An area 1 is obtained, a partial area 2 which is an overlapping portion of the current area 2 and the previous area 2 is obtained, and the current area 1 and the current area 2 are obtained.
Partial area processing function for obtaining a partial sum area which is an overlapping part of the sum area generated by the sum area and the previous sum area, and a partial area which is the sum of the pixel values of the partial areas 1, 2 and 2. A partial area sum calculating function for calculating the sum, a partial area square sum calculating function for calculating a partial area square sum that is a sum of squares of the pixel values of the partial area 1, the partial area 2, and the partial sum area; From the previous time sum that is the sum of the pixel values of the area 1, the area 2 and the sum area, and the sum of the partial areas, the current sum that is the sum of the pixel values of the area 1, the area 2 and the sum area at the present time is obtained. From the current time sum calculation function, the previous time sum of squares, which is the sum of the squares of the pixel values of the area 1, the area 2, and the sum area at the previous time, and the partial area sum of squares,
A current square sum calculation function for calculating a current square sum that is a sum of squares of pixel values of the area 2 and the sum area, and a separation degree calculation function for calculating a separation degree from the current sum and the current square sum are realized. Recording medium storing a program for a filtering process of the degree of separation, characterized by recording a program for the separation.

According to a tenth aspect of the present invention, the partial area sum and the partial area square sum are calculated from the area 1 and the area 2 in the image, and the degree of separation is further calculated from the partial area sum and the partial area square sum. A large frame-shaped area obtained by superimposing a middle rectangular area on a large rectangular area is set as an area 1 in a recording medium on which a program for realizing a degree of separation filtering process for extracting feature points in the image is recorded. An area 1 setting function, an area 2 setting function of setting a frame-shaped small area obtained by superimposing a rectangular small area in a rectangular middle area as an area 2, an area 1, an area 2 and a sum of the areas 1 and 2 A sum calculation function for calculating the sum of the pixel values of the areas from the sum of the large rectangular area, the middle rectangular area, and the small rectangular area, and the sum of the squares of the pixel values of the areas 1, 2, and the sum of the square areas. Sum of squares of area and rectangular small area A program for realizing a sum-of-squares calculation function to calculate the degree of separation and a separation degree calculation function of calculating the degree of separation from the sum of the pixel values of the areas 1, 2, and the sum area and the sum of squares is recorded. This is a recording medium on which a program for a degree of separation filtering process is recorded.

According to an eleventh aspect of the present invention, the sum calculation function performs the calculation of the sum of the pixel values of each rectangular area separately in horizontal and vertical processes.
11. The method according to claim 10, wherein the calculation of the sum of squares of the pixel values of each rectangular region is performed separately in horizontal and vertical directions.
It is a recording medium on which a program for the degree of separation filtering process described above is recorded.

According to a twelfth aspect of the present invention, a horizontal partial area 1 which is an overlapping portion of a current horizontal area 1 and a previous horizontal area 1 which is a time earlier than the current time is obtained, Region 2 and previous horizontal region 2
Is obtained, and a horizontal portion which is an overlapping portion of the horizontal sum region generated from the current horizontal region 1 and the horizontal direction region 2 and the previous horizontal sum region is obtained. It has a horizontal partial area processing function for obtaining a sum area, and obtains a vertical partial area 1 which is an overlapping part of the current vertical area 1 and the previous vertical area 1 which is a time point before the current time. A vertical partial area 2 which is an overlapping part of the current vertical area 2 and the previous vertical area 2 is obtained, and the current vertical area 1 is obtained.
And a vertical partial area processing function for obtaining a vertical partial sum area that is an overlapping part of the vertical sum area generated from the vertical area 2 and the previous vertical sum area. A partial area sum calculation function for calculating a partial area sum that is the sum of the pixel values of the partial areas 1, 2 and the partial sum area for each of the horizontal direction and the vertical direction; From the sum at the previous point in time, which is the sum of the pixel values of the sum area, and the sum of the partial areas, the current area 1,
A current sum calculation function for calculating a current sum that is a sum of pixel values of the area 2 and the sum area. The square sum calculation function includes a partial area 1 and a partial area 2 for each of the horizontal direction and the vertical direction.
And a partial area square sum calculation function for calculating a partial area square sum that is a square sum of the pixel values of the partial sum area, and a square sum of the pixel values of the area 1, the area 2, and the sum area at the previous time. One point before 2
It has a current square sum calculation function for calculating a current square sum that is a sum of squares of pixel values of the current region 1, the region 2, and the sum region from the sum of squares and the sum of squares of partial regions. A recording medium storing a program for a degree-of-separation filtering process according to claim 11.

[0018]

BEST MODE FOR CARRYING OUT THE INVENTION

(Outline of Separation Filtering) First, the separation filtering which is a premise of the present invention will be briefly described.

The "separation degree" is a measure indicating the difference in the statistical properties of images in two certain regions on the image, and is mainly used for detecting the contours and feature points of the image.

The degree of separation η is defined by equation (1).

[0021]

(Equation 1) Each code in the expression (1) is defined as follows for two arbitrary regions on the image.

[0022] _{N 1} the number of pixels in the region 1, _{N 2} the number of pixels in the region 2, _{N a} number of pixels in the entire area, the dispersion value of the entire sigma _a ² region, f (i) the position i luminance level, the average luminance level of an average f ₁ region 1, the average luminance level of an average f ₂ region 2, the average f _m average luminance level of the entire area in.

Note that the entire area is the sum of the area 1 and the area 2 and does not represent the entire image area (the entire image area is described as the entire image area).

The degree of separation η has a value in the range of 0 <η ≦ 1;
It becomes 1 when the information of the arbitrary areas 1 and 2 on the image is completely separated.

Since the degree of separation is a measure indicating the difference in the statistical properties of the images in the two regions, it is hardly affected by noise. Effective for extraction.

FIG. 1 shows an example of how to select the regions 1 and 2.

FIG. 1A shows a case where two areas are one-dimensionally selected and contour detection is performed. References include Kazuhiro Fukui, "Contour detection using separation between regions",
JIJI-Kyoho CV88-2, pp.9-16,1994.Yamaguchi, Fukui, "Face Image Analysis Using Separation Features-Eye Pupil Detection", The 52nd Annual Conference of the Japan Society of Jisho (2) -pp. .187-188, 1996. At this time,
The degree of separation increases in the image outline portion (vertical outline in the mask in the figure), and the image has a local portion along the outline.

FIG. 1B shows an example used in Japanese Patent Application No. 9-33704, in which an area is selected two-dimensionally. In this example, a mask used for detecting a circular shape (hereinafter, referred to as a mask)
"Separation degree mask"). The degree of separation has a local maximum value in a circular portion such as the pupil. Therefore, a point (pixel coordinate) where the degree of separation has a local maximum value is a feature point candidate. It should be noted that such processing using the separation mask is called separation filtering.

In the above example, the image brightness level is used. However, the present invention can be applied to other image features such as hue and saturation instead. Furthermore, a feature amount completely different from an image such as depth information obtained from a sonar radar may be used.

(Embodiment 1) The problem in performing the filtering of the degree of separation is, as can be seen from the definition equation of the degree of separation, the high computational cost in the case of using a large area processing or a large degree of separation mask. is there. If the operation cost can be reduced, the effect is very large because the circuit scale of the processing device is reduced, and conversely, the processing at a high frame rate and the selection of the isolation mask are widened. An object of the present invention is to reduce the computation cost of the separation filtering process as compared with the related art, and assume that the separation filtering is performed on a relatively large area in the image area, thereby eliminating redundancy in the computation process. , The amount of calculation is reduced.

The details of the present invention will be described below.

However, the present invention is mainly applicable to the case shown in FIG.
This embodiment is particularly effective when filtering the degree of separation in a large area of an image using a two-dimensional mask, and the embodiment will be described with reference to a two-dimensional mask.

The separation filtering process described below is hereinafter referred to as a "sequential separation filtering process".

First, the definition equation of the degree of separation is rewritten. However,
Note that the notation is different. Point X = (x,
When calculating the degree of separation in y), the area 1 and the area 2 used for this calculation are R ₁ (X) and R ₂ (X), respectively.
Note that R ₁ and R ₂ are functions. This indicates that R ₁ (X) and R ₂ (X) are regions 1 and 2 used for calculating the degree of separation at position X.

[0035]

(Equation 2) It is.

[0036] _However, S 1, _{S 2} is the sum of the pixel values of the region 1, 2, _{S a} total sum of pixel values in the entire area, _{T a} is the sum of the squares of the pixel values of the entire region, _{N 1} , _N 2, _{N a} are each region 1,
Region 2 is the number of pixels in all regions.

Next, consider a case where the filtering of the degree of separation is performed within a certain area in the image.

The pixel points in the area where the degree of separation is calculated (in the filtering area) are represented by X _i : i = 1, 2,... N.

In the process of calculating the degree of separation, equations (5) to (8)
Therefore, it is necessary to calculate the sum or the sum of squares in the isolation mask area. 2, X of the previous time and a certain pixel point X _i
It shows the relationship of the area 1 of the _i-1 isolation mask. As is clear from this figure, when there is an area shared between the areas, there is redundancy in the process of calculating the sum or the sum of squares in the area 1, and the overlapping calculation of the common area is performed. It turns out that it is possible to omit. Therefore,
By omitting overlapping calculations in the shared area, the amount of calculation can be significantly reduced. This discussion is of course valid for the region 2 and the entire region (the sum of the region 1 and the region 2).

Hereinafter, this calculation process will be described.

[0041]

(Equation 3) Therefore, if there is a common area between each of the isolation mask areas at the two points X _i−1 and X _i on the image on which the isolation filtering is performed, the summation and the two-summation calculation are serialized as in the above equation. By doing so, the amount of calculation can be reduced, and the speed of the separation filtering process can be increased.

FIG. 3 is a flowchart showing the entire flow of the sequential type separation degree filtering process. FIG.
FIG. 2 is a block diagram of a separation degree filtering device 10 for performing a sequential separation degree filtering process.

First, the initialization processing section 11 calculates the sum or the sum of squares of the areas 1 and 2 and the whole area at the first point where the separation filtering is performed by the initialization processing (step 1-1). . This calculation result is the initial value of the recurrence formula. Also, initialization of a temporary variable for determining the end of calculation (when i = 0 is the initial position, i =
(Initialize to 1).

Next, in the partial area sum / square sum calculation processing (step 1-2), the partial area sum / square sum calculation processing unit 12 calculates the areas 1 and 2 at time (i-1) and time i. The sum and the square sum are calculated for the partial areas generated by the relationship

In step 1-3, the summation / square sum calculation result in the initialization processing (step 1-1) or i-1 and the sum / square sum calculation in the partial area obtained in step 1-2 Using the result, the degree of separation at the time point i is calculated by the degree of separation calculation processing unit 13.

In step 1-4, the control section 14 adds a temporary variable to determine the end of the processing, and if not, returns to step 1-2 to continue the processing.

FIG. 5 is a flowchart showing a detailed example of the above processing.

First, in step 2-1, S ₁ (X
₀ ), S ₂ (X ₀ ), S _a (X ₀ ), and T _a (X ₀ ) are calculated by equations (5), (6), (7), and (8), respectively.

Next, a variable i for ordering the point sequence on the image
Is initialized to 1 (step 2-2).

In step 2-3, the second and third terms of equation (9) are calculated.

Similarly, in step 2-4, the second and third terms of equation (11) are calculated in step 2-5 of equation (10), and in step 2-6, equation (12) is calculated.

In step 2-7, if i = 1, step 2-1 is performed. If i is not 1, S ₁ (X _i-1 ) and S ₁ calculated in the previous step 2-7 are calculated. ₂ (X
_{i- 1), S a (X} i-1), using the results of T a _{(X i-1)} and step 2-3～2-6 formula (9) to (1
2), S ₁ (X _i ), S ₂ (X _i ), S
_{a (X i),} to calculate the T a _{(X i).}

In step 2-9, S ₁ (X _i ), S ₂ (X _i ), S _a (X _i ), and T _a obtained in step 2-7.
(X _i ) is substituted into Expressions (2) to (4) to obtain the degree of separation η.
(X _i ) is obtained.

Next, i is incremented (step 2-9).

Finally, if i <n, step 2
The process is shifted to -3, and when i = n, the process ends.

[0056]

(Equation 4) FIG. 6 is an example of a flowchart showing the flow of processing when there is no shared area between the areas R ₁ and R ₂ .

In this example, only some steps are changed from the case where there is a shared area (FIG. 5).
Here, only this change will be described.

[0058] First, of the processing of the processing step 2-7 in FIG. 5, the _{S a} for no shared area (11)
Regardless of this, the expression (13) is changed. Further, since the processing corresponding to step 2-5 is not necessary, this processing is deleted.

(Embodiment 2) Next, a second embodiment will be described.

In this embodiment, the amount of calculation is reduced by setting the isolation mask to a rectangular block as shown in FIG. 7 and setting the areas 1, 2, and 3 so as not to have any areas shared by each other. . Hereinafter, the details of the method for reducing the amount of computation will be described. Note that the finally obtained degree of separation relates to the region 1 and the region 2 as in the first embodiment. The content of the separation degree filtering device 10 is the same as that of FIG.

Further, several conditions are assumed for the sake of simplicity. First, a region (filtering region) in which separation filtering is performed is set as a rectangular region in an image. Although only the example of calculating the degree of separation in this area in dictionary order is taken up, in general, it can be easily extended to other than this scanning method. Further assume that the image is large enough. This assumption is for eliminating an exceptional process at the image boundary.
However, these assumptions are for facilitating the explanation and the expansion of equations, and do not impose any restrictions on the algorithm described below.

First, this mask is divided into three regions as shown in FIG. The following relationship holds between the two masks.

[0063]

(Equation 5) At this time, the sum or square sum in both regions has a relationship as shown in FIG. That is, P ₁ , P ₂ , P
₃ is the sum of the pixel values of the regions R ′ ₁ , R ′ ₂ , R ′ ₃ respectively;
Q _1, if the sum of the squares of the _{Q 3} each region R _'1, R' pixel values in the _{3, S 1, S 2,} S a, T a and _{P 1, P} 2, P
₃ , Q ₁ and Q ₃ have the following relationship.

[0064]

(Equation 6) Therefore, instead of the sum or square sum in the region _R 1, _{R 2} directly, region _{R '1, R' 2,} R ' may be obtained them from the sum or square sum of _3. There are other equivalent equations, but equations (16) to (19) are used here.

By the way, when calculating the sum or the sum of the squares in the rectangular area (that is, R ′ ₁ , R ′ ₂ , R ′ ₃ ) at each point in the separation degree calculation area, as shown in FIG.
It is possible to separate the processing in the horizontal direction and the processing in the vertical direction. By separating the processing in the horizontal direction and the vertical direction, the amount of calculation can be reduced. However, in addition to the sequential type separation degree filtering processing as described in the first embodiment, the calculation amount is further reduced. Reduce.

FIG. 10 shows an example of processing in the vertical direction.

First, at the first point (y), y + l _yi to y
Calculate the sum up to + u _yi . Where y + l _yi , y + u
_yi is a constant indicating the size of the rectangular area (FIG. 8). At the next point (y + 1), the calculation result at y and y + _liy , y
Using the pixel value at + u _iy, the sum at y + 1 is calculated. Therefore, by sequentially performing this processing, the amount of calculation can be reduced.

[0068] In the _{P 1} as an example, illustrates this process by the equation. First,

(Equation 7) It is. l _1x , l _1y , u _1x , and u _1y are constants representing the size of the region as shown in FIG. 7 (in the figure, the vertical coordinate becomes larger as going downward). In the case of this example, equation (24)
In sequentially obtains the vertical summation, equation (21) As a result of the horizontal sum (ie sum of area R _'1 of the pixel value)
Are sequentially obtained.

Here, the conventional method is to perform the calculation using the equation (20), and the method of the present invention is to perform the calculation using the equations (22) and (24). The difference is the difference in the calculation speed.

Similarly, P ₂ , P ₃ , Q ₁ and Q ₂ are as follows.

[0071]

(Equation 8) Next, an example of a specific processing flow will be described.
Assume that the filtering area is as shown in FIG. However,
It is assumed that no exceptional processing occurs near the image boundary as described above.

Description will be made by taking P ' ₁ and P ₁ as examples. Since it is necessary to obtain the initial value in the sequential calculation, it is necessary to change the calculation method depending on the calculation point. Also, from the equation (22) takes the larger than a pre-filtering region an area for calculating the ₁ 'P for ₁ requires' filtering region outside of P when determining the P _1.

[0073] (a) region _{(x 0 + l x, y} o) - (x 1 + u x, y 0) calculated using equation (23) (b) area _{(x 0 + l x, y} o +1) - ( x ₁ + _u x, y
₁ ) Calculation using Expression (24) Here, the process (a) is an initial value calculation process. Therefore, after processing (a), process the P _'1 from performing the (b) determining.

Next, determine the _{P 1} using the calculation result.
Also in this case, it is necessary to change the calculation method depending on the calculation point by the initial value calculation.

(C) Processing area (x ₀ , y _o ) − (x ₁ +1, y ₁ ) Calculated by equation (21) (d) Processing area (x ₀ +1, y _o ) − (x ₁ , y ₁₎ In this case, the process (c) is the initial value process. Accordingly, after the process (c), P ₁ is obtained in each calculation point by performing processing (d).

Similarly, P ′ ₂ , P ′ ₃ , Q ′ ₁ , and Q ′ ₂ can be calculated. Therefore, _S 1 by the equation _{(16) ~ (19),} S 2, S a, T a can be calculated, equation (2),
From (3) and (4), the degree of separation can be obtained.

It should be noted that the separation degree filtering device 1 for performing the sequential separation degree filtering processing of the first and second embodiments.
0 can be realized by a personal computer, and a program for a function of realizing the sequential separation degree filtering processing is recorded in an FD or a CD-ROM.

[0078]

According to the present invention, as can be seen from the definition formula of the degree of separation in performing the degree of separation filtering, the operation cost in the case of using a large area processing or a large degree of separation mask can be reduced. The circuit scale of the processing device can be reduced, and the range of processing at a high frame rate and the selection of a separation mask can be widened.

[Brief description of the drawings]

FIG. 1 is an example of a region shape used for calculation of a degree of separation (separation degree mask example 1).

FIG. 2 is a relationship between regions.

FIG. 3 is an overall flow of separation degree filtering.

FIG. 4 is a block diagram of a separation degree filtering device.

FIG. 5 is a first example of separation filtering processing;

FIG. 6 is a second example of the separation filtering process.

FIG. 7 is an example of a rectangular separation mask (separation mask example 2).

FIG. 8 illustrates division of a rectangular separation mask.

FIG. 9 shows a relationship between a division mask and a separation mask.

FIG. 10 is an example of separation between horizontal processing and vertical processing.

FIG. 11 is an example of sequential calculation in the vertical direction.

FIG. 12 is a filtering area.

[Explanation of symbols]

 REFERENCE SIGNS LIST 10 separation degree filtering device 11 initialization processing part 12 partial area sum / square sum calculation processing part 13 separation degree calculation processing part 14 control part

Claims (12)

[Claims] 1. A partial area sum and a partial area square sum are calculated from an area 1 and an area 2 in an image, and the degree of separation is further calculated from the partial area sum and the partial area square sum. In a separation degree filtering processing apparatus for extracting a feature point, a partial area 1 which is an overlapping portion of a current area 1 and a previous time area 1 which is a time point earlier than the current time area is obtained, and a current area 2 and a previous time area 2 A partial area processing means for determining a partial area 2 which is an overlapping part of the area 2 and obtaining a partial sum area which is an overlapping part of the sum area generated from the current area 1 and the area 2 and the previous sum area; A partial area sum calculating means for calculating a sum of partial areas which is a sum of pixel values of the partial areas 1, 2 and the sum of squares of pixel values of the partial areas 1, 2 and 2. Partial area 2 which is A partial area square sum calculating means for calculating the sum; a previous time total which is the sum of the pixel values of the area 1, the area 2 and the sum area at the previous time; Current sum calculation means for calculating a current sum total which is a sum of pixel values of the sum 2 and the sum area; a previous sum of squares which is a sum of squares of the pixel values of the area 1, the area 2 and the sum area at the previous time; Current square sum calculating means for calculating the current square sum which is the sum of the squares of the pixel values of the current area 1, the area 2 and the sum area from the area sum of squares; A separation degree filtering processing apparatus, comprising: a separation degree calculation unit for calculating a separation degree. 2. A partial area sum and a partial area square sum are calculated from the areas 1 and 2 in the image, and a degree of separation is further calculated from the partial area sum and the partial area square sum. A separation degree filtering processing apparatus for extracting a feature point, comprising: a region 1 setting means for setting a frame-shaped large region obtained by superimposing a rectangular medium region on a rectangular large region as a region 1; Area 2 setting means for setting a frame-shaped small area obtained by superimposing small areas as area 2; area 1, area 2, and the sum of pixel values of the sum area of area 1 and area 2, A sum calculating means for calculating the sum of the pixel values of the area 1, the area 2, and the sum area from the sum of the squares of the large rectangular area, the medium rectangular area, and the small rectangular area;
A separation filtering processing apparatus, comprising: a sum of squares calculating means; and a separation degree calculating means for calculating the degree of separation from the sum of the pixel values of the area 1, the area 2, and the sum area and the sum of squares. 3. The method according to claim 1, wherein the sum calculating means calculates the sum of the pixel values of each rectangular area separately in horizontal and vertical processing. 3. The apparatus for filtering the degree of separation according to claim 2, wherein the calculation of the sum of squares is performed separately in horizontal and vertical directions. 4. A horizontal partial area 1 which is an overlapping portion between the current horizontal area 1 and the previous horizontal area 1 which is a time before the current time is obtained, and the current horizontal area 2 and the previous time The horizontal part area 2 which is an overlapping part of the horizontal area 2 is obtained, and the overlapping part of the horizontal sum area generated from the current horizontal areas 1 and 2 and the previous horizontal sum area A horizontal partial area processing means for obtaining a horizontal partial sum area which is a vertical partial area which is an overlapping part of a current vertical area 1 and a previous vertical area 1 which is a time point before the current vertical area 1 1 is obtained, a vertical partial area 2 which is an overlapping portion of the current vertical area 2 and the previous vertical area 2 is obtained, and the current vertical area 1 and the current vertical area 2 are obtained.
Vertical partial area processing means for obtaining a vertical partial sum area which is an overlapping part of the vertical sum area generated from the previous vertical sum area and the vertical sum area at the previous point in time. A partial area sum calculating means for calculating a sum of partial areas which is a sum of pixel values of the partial areas 1, 2 and the sum of pixel values of the areas 1, 2 and the sum area at the previous time. And a current sum calculating means for calculating a current sum that is a sum of pixel values of the current area 1, the area 2 and the sum area from the previous time sum and the partial area sum. Means for calculating a partial area sum of squares which is a sum of the squares of the pixel values of the partial area 1, the partial area 2 and the partial sum area for each of the horizontal direction and the vertical direction; Of the pixel values of the region 1, the region 2 and the sum region Current sum of squares calculation means for obtaining the current sum of squares, which is the sum of the squares of the pixel values of the current region 1, the region 2, and the sum region, from the previous sum of squares of the previous time and the sum of the squares of the partial regions. The filtering apparatus according to claim 3, further comprising: 5. A partial area sum and a partial area square sum are calculated from the areas 1 and 2 in the image, and the degree of separation is further calculated from the partial area sum and the partial area square sum. In the separation degree filtering processing method for extracting a feature point, a partial area 1 which is an overlapping portion of a current area 1 and a previous time area 1 which is a time point earlier than the current time area is obtained, and a current area 2 and a previous time area 2 A partial area processing step of obtaining a partial area 2 which is an overlapping part of the area 2 and obtaining a partial sum area which is an overlapping part of the sum area generated from the current area 1 and the area 2 and the previous sum area; Calculating a partial area sum that is a sum of pixel values of the partial area 1, the partial area 2, and the partial sum area; and a square sum of pixel values of the partial area 1, the partial area 2, and the partial sum area Department A partial area square sum calculating step of calculating the area square sum, a previous time total sum of pixel values of the area 1, the area 2 and the sum area at the previous time, and a partial area total, 1, a current sum calculation step for calculating a current sum that is a sum of pixel values of the area 2 and the sum area; and a previous sum of squares which is a sum of squares of pixel values of the area 1, the area 2 and the sum area at the previous time A current square sum calculation step of calculating a current square sum that is a sum of squares of pixel values of the current region 1, the region 2, and the sum region from the partial region square sum, a current total sum and a current square And a separation degree calculating step of calculating the separation degree from the sum. 6. A partial area sum and a partial area square sum are calculated from the areas 1 and 2 in the image, and the degree of separation is further calculated from the partial area sum and the partial area square sum. A separation degree filtering method for extracting feature points, comprising: a region 1 setting step of setting as a region 1 a frame-shaped large region obtained by superimposing a rectangular middle region on a rectangular large region; An area 2 setting step of setting a frame-shaped small area obtained by superimposing the small areas as an area 2, and a sum of pixel values of an area 1, an area 2, and a sum area of the area 1 and the area 2, A sum calculation step of calculating from the sum of the area and the small rectangular area; and a square calculation of the square value of the pixel values of the area 1, the area 2, and the sum area from the square sum of the large rectangular area, the middle rectangular area, and the small rectangular area.
A separation degree filtering processing method, comprising: a sum of squares calculation step; and a separation degree calculation step of calculating a separation degree from the sum of the pixel values of the area 1, the area 2, and the sum area and the sum of squares. 7. The step of calculating the total sum of the pixel values of each rectangular area is performed separately in horizontal and vertical directions, and the step of calculating the sum of squares includes calculating the pixel value of each rectangular area. 7. The method according to claim 6, wherein the calculation of the sum of the squares is performed separately in horizontal and vertical directions. 8. A horizontal partial area 1 which is an overlapping portion between the current horizontal area 1 and the previous horizontal area 1 which is a time point before the current time is obtained, and the current horizontal area 2 and the previous time point are obtained. The horizontal part area 2 which is an overlapping part of the horizontal area 2 is obtained, and the overlapping part of the horizontal sum area generated from the current horizontal areas 1 and 2 and the previous horizontal sum area And a horizontal partial area processing step for obtaining a horizontal partial sum area that is a vertical partial area that is the overlapping part of the current vertical area 1 and the previous vertical area 1 that is a time point before the current time. A partial area 1 is obtained, a vertical partial area 2 which is an overlapping portion of the current vertical area 2 and the previous vertical area 2 is obtained, and the current vertical area 1 and the current vertical area 2 are obtained.
A vertical partial area processing step of obtaining a vertical partial sum area, which is an overlapping part of the vertical sum area generated from the previous vertical sum area and the vertical sum area at the previous point in time. For each direction, a partial area sum calculating step of calculating a partial area sum that is the sum of the pixel values of the partial areas 1, 2 and the pixel values of the areas 1, 2 and the sum area at the previous time A current sum calculation step of calculating a current sum that is a sum of pixel values of the current region 1, the region 2, and the sum region from the previous time sum that is the sum of The calculation step includes, for each of the horizontal direction and the vertical direction, a partial area square sum calculating step of calculating a partial area square sum that is a square sum of pixel values of the partial area 1, the partial area 2, and the partial sum area; Region 1 at the time, From the previous sum of squares of the pixel values of the area 2 and the sum area and the sum of the squares of the partial area, the current time which is the sum of the squares of the pixel values of the area 1, the area 2 and the sum area at the present time 8. The method of claim 7, further comprising a current-sum-of-squares calculation step of obtaining a sum of squares. 9. A partial area sum and a partial area sum are calculated from the areas 1 and 2 in the image, and the degree of separation is further calculated from the partial area sum and the partial area sum. On a recording medium storing a program for realizing a degree-of-separation filtering process for extracting a feature point, a partial area 1 which is an overlapping portion of the current area 1 and the previous time area 1 which is a time point before the current time is obtained. A partial area 2 which is an overlapping portion between the current area 2 and the previous time area 2 is obtained, and a partial sum area which is an overlapping portion between the current area 1 and the sum area generated from the area 2 and the previous time sum area. A partial area summation function for calculating a partial area sum that is the sum of the pixel values of the partial areas 1, 2 and the partial area 1, a partial area 2, and a partial sum Territory A partial area sum-of-squares calculation function for calculating a partial area sum-of-squares which is a sum of squares of the pixel values of: a sum of pixel values of the area 1, the area 2 and the sum area at the previous time; A current sum calculation function for calculating a current sum that is the sum of the pixel values of the current region 1, the region 2 and the sum region from the partial region sum; and a current sum calculation function for calculating the pixel values of the region 1, the region 2 and the sum region at the previous time. From the previous sum of squares, which is the sum of squares, and the sum of squares of the partial area, the current sum of squares, which is the sum of the squares of the pixel values of the current area 1, the area 2, and the sum area, is obtained. A recording medium storing a program for a separation degree filtering process, wherein a program for realizing a calculation function and a separation degree calculation function for calculating a separation degree from a current sum and a current square sum is recorded. 10. A partial area sum and a partial area square sum are calculated from the area 1 and the area 2 in the image, and the degree of separation is further calculated from the partial area sum and the partial area square sum. An area 1 setting function for setting a frame-shaped large area obtained by superimposing a rectangular middle area on a rectangular large area as an area 1 in a recording medium storing a program for realizing a separation degree filtering process for extracting a feature point; An area 2 setting function for setting a frame-shaped small area obtained by superimposing a rectangular small area on a middle rectangular area as an area 2, and an area 1, an area 2 and a pixel value of a sum area of the area 1 and the area 2 A sum calculation function for calculating the sum from the sum of the large rectangular area, the medium rectangular area, and the small rectangular area; and the square sum of the pixel values of the areas 1, 2, and the large area, the medium rectangular area, and the small rectangular area. 2 calculated from the sum of squares of
A separation degree characterized by recording a program for realizing a sum of squares calculation function and a separation degree calculation function of calculating a separation degree from a sum of pixel values of a region 1, a region 2 and a sum region and a square sum. A recording medium on which a program for filtering processing is recorded. 11. The summation function performs the calculation of the sum of the pixel values of each rectangular area separately in horizontal and vertical processing, and the square sum calculation function calculates the pixel value of each rectangular area. 11. The recording medium according to claim 10, wherein the calculation of the sum of squares is performed separately in horizontal and vertical directions. 12. A horizontal partial area 1 which is an overlapping portion of a current horizontal area 1 and a previous horizontal area 1 which is a time point before the current time is obtained, and a current horizontal area 2 and a previous time point are obtained. The horizontal part area 2 which is an overlapping part of the horizontal area 2 is obtained, and the overlapping part of the horizontal sum area generated from the current horizontal areas 1 and 2 and the previous horizontal sum area And has a horizontal partial area processing function for obtaining a horizontal partial sum area that is a vertical partial sum area, and a vertical direction that is an overlapping part of the current vertical area 1 and the previous vertical area 1 that is a time point before the current time. A partial area 1 is obtained, a vertical partial area 2 which is an overlapping portion of the current vertical area 2 and the previous vertical area 2 is obtained, and the current vertical area 1 and the current vertical area 2 are obtained.
A vertical partial area processing function for obtaining a vertical partial sum area that is an overlapping part of the vertical sum area generated from the previous vertical sum area and a vertical sum area at a previous time point. For each direction, a partial area sum calculation function for calculating the sum of the partial areas, which is the sum of the pixel values of the partial areas 1, 2 and the pixel values of the areas 1, 2 and the sum area at the previous time A total sum of pixel values of the current region 1, the region 2 and the sum region, and a current total sum calculating function for calculating a current total sum of pixel values of the current region 1, the region 2 and the sum region from the previous total sum of The calculation function includes, for each of the horizontal direction and the vertical direction, a partial area square sum calculation function for calculating a partial area square sum that is a sum of squares of pixel values of the partial area 1, the partial area 2, and the partial sum area; Pixel values of region 1, region 2, and sum region at the time From the previous sum of squares, which is the sum of squares, and the sum of squares of the partial area, the current sum of squares, which is the sum of the squares of the pixel values of the current area 1, the area 2, and the sum area, is obtained. 12. A recording medium storing a program for a degree of separation filtering process according to claim 11, which has a calculation function.