JP4975861B2 - Edge direction detection apparatus or method - Google Patents

Description

  The present invention relates to an edge direction detection device, and more particularly to edge direction detection capable of highly accurate edge detection.

  FIG. 1 shows the RGB values of each pixel when there is a contour on the screen. For use in image processing such as image interpolation in super-resolution and enlarged display, edge detection with high accuracy in such an image is required. Here, “edge” refers to a region and a region in which the above characteristics change suddenly when a region having similar features such as luminance value, density value, color, and pattern of pixels is defined as one region. And the boundary.

  In Patent Document 1, in order to improve accuracy, a virtual pixel is set at the center of a window of 18 pixels (2 * 9), and edge detection is performed by obtaining a correlation measurement value from the sum of differences between upper and lower corresponding pixels. An edge detection technique is disclosed.

  Specifically, as shown in FIG. 2, a virtual pixel (i, j) is set in the center of an 18 pixel (2 * 9) window. The correlation measurement value DIFF_MID in the vertical direction can be defined by the following equation (1). In equation (1), the sum total of K = −7 to 7 is obtained, so the sum total of 15 combinations obtained by shifting the window of (2 * 9) by 7 to the left and right is obtained.

Here, L (y, x) is a pixel value of the pixel (y, x).

  Similarly, DIFF_L2 to DIFF_L9 and DIFF_R2 to DIFF_R9 are obtained for diagonal pixel combinations as shown in FIG.

  For example, DIFF_R7 and DIFF_L7 are obtained by Expression (2) and Expression (3).

  Edge detection is performed based on the correlation measurement value.

  Patent Document 2 discloses a technique capable of edge detection in 12 directions. Specifically, as shown in FIG. 4, a part of a 6 * 6 rectangular window is removed, and virtual interpolation pixels G0 to G4 are set for a substantially diamond-shaped window composed of 24 pixels, A plurality of virtual dividing lines L passing through the interpolation pixel G0 are set at a pitch of 12.5 degrees. Specifically, 12 dividing lines are set. The total of the pixels on both sides divided by these dividing lines is obtained in each of 12 directions, and it is determined that the direction of the dividing line where the value obtained by multiplying the absolute value of the total difference by the coefficient is the edge direction.

JP 2004-201308 JP JP 2009-94862 A

  The edge detection method disclosed in the above patent document has the following problems.

  In Patent Document 1, since the edge detection is performed by the calculation formula of the measurement value, not only the calculation for two lines of 9 + 2 * 7 = 23 pixels is actually required, but also the vicinity of the vertical (from 45 degrees to 45 degrees) In order to detect an edge with the same accuracy for between 135 degrees, a memory having the same number of lines is required, and an arithmetic process for that is required.

  In Patent Document 2, since the determination is based on the absolute difference of the sum of the values of the pixels separated by the virtual edge, the necessary pixel area is wide and the edge detection is not stable particularly in a complicated portion or checkered pattern. .

  SUMMARY OF THE INVENTION An object of the present invention is to provide an edge detection apparatus or method that solves the above-described problems and that can detect edges with high accuracy without being bound by the edge direction.

(1) An edge direction detection apparatus according to the present invention calculates edge detection reliability for a predetermined edge direction candidate in a rectangular area of M pixels * N pixels, and based on the edge detection reliability, the edge direction in the rectangular area A) an edge direction detecting device for detecting A, wherein one pixel is an adjacent pixel adjacent to a central pixel of the rectangular region, and the other pixel forms a line segment with the adjacent pixel. Combination pixel storage means for storing at least a plurality of combinations in which the line segments are in parallel with respect to each edge direction, and B) for the designated rectangular area, the absolute value of the difference between the combination pixels in the combination pixel storage means, A difference total value calculation means for calculating a difference total value totaled for each edge direction for each edge direction; C) the combination pixel , A central neighborhood pixel value calculating means for calculating a representative pixel value in each edge direction based on a pixel value of an adjacent pixel adjacent to the central pixel of the rectangular area, and D) each pixel constituting the designated rectangular area E) a representative pixel value calculating means for calculating a representative pixel value in the rectangular area from the value of E, a center for calculating a difference between the central neighboring pixel value and the representative pixel value as a central neighboring pixel difference value for each edge direction Neighboring pixel difference value calculation means, F) For each edge direction, for each edge direction, a value obtained by multiplying the difference total value by a first coefficient, and a value obtained by multiplying the center neighboring pixel difference value by a second coefficient. G) edge detection reliability calculation means for calculating a higher edge detection reliability as the sum of the two is smaller , and G) determining that the edge direction having the higher edge detection reliability is the edge direction in the rectangular area. Determining means to perform.

Thus, for each edge direction, the difference between the central neighborhood pixel value and the representative pixel value is calculated as a central neighborhood pixel difference value, and for each edge direction , the first coefficient is added to the difference total value. By calculating a higher edge detection reliability as the sum of the multiplied value and the value obtained by multiplying the second neighboring coefficient by the center neighboring pixel difference value is smaller, edge detection with higher reliability becomes possible.

(2) In the edge direction detection apparatus according to the present invention, the first and second coefficients are 1.
(3) In the edge direction detection apparatus according to the present invention, the edge direction stored in the combination pixel storage unit is a direction parallel to a line segment that connects the existing pixels in the rectangular region and passes through a rectangular central pixel. is there. Accordingly, it is possible to detect the edge direction in the interpolation virtual pixel exceeding the actual pixel unit.

( 4 ) In the edge direction detection apparatus according to the present invention, the number of combinations of pixels in each edge direction is at least 3, and the weighting coefficient at the center is increased. Accordingly, a combination of a plurality of pixels is used, and an edge based on a pixel near the center is detected.

( 5 ) The edge direction detection device according to the present invention calculates edge detection reliability for a predetermined edge direction candidate in a rectangular area of M pixels * N pixels, and based on the edge detection reliability, the edge direction in the rectangular area A) an edge direction detecting device for detecting A, wherein one pixel is an adjacent pixel adjacent to a central pixel of the rectangular region, and the other pixel forms a line segment with the adjacent pixel. Combination pixel storage means for storing at least a plurality of combinations in which the line segments are in parallel with respect to each edge direction, and B) for the designated rectangular area, the absolute value of the difference between the combination pixels in the combination pixel storage means, Difference total value calculation means for calculating the difference total value totaled for each edge direction for each edge direction ; C) for each edge direction Edge detection reliability calculation means for calculating a higher edge detection reliability as the difference total value is smaller , and D) a determination means for determining that the edge direction having the higher edge detection reliability is the edge direction in the rectangular area. E) The number of combinations of pixels in each edge direction stored in the combination pixel storage means is at least 3, and the total weighting coefficient is increased when totaling.

  In this way, an edge based on a pixel near the center is detected while using a combination of a plurality of pixels.

( 6 ) The edge direction detection device according to the present invention obtains a horizontal difference and a vertical difference in the rectangular area, respectively, and if the horizontal minimum value is smaller than a threshold value, the vertical direction is not vertical. When the minimum value of the direction is larger than the threshold value, it is determined that the direction is not horizontal. Therefore, it is possible to prevent erroneous detection due to a combination of pixels constituting the vertical direction or the horizontal direction.

(7) The edge direction detection device according to the present invention provides a difference between a combination of adjacent pixels in the positive direction connecting the first quadrant and the third quadrant or the negative direction connecting the second quadrant and the fourth quadrant in the rectangular area. Are obtained in the positive direction and the negative direction, and are determined not to be edge directions belonging to the larger direction. Therefore, erroneous detection can be prevented.

(8) In the edge direction detection device according to the present invention, the rectangular area is a square, and the combination pixel storage unit is configured so that, when the edge direction is parallel to the diagonal of the square, A combination located on the side and a combination located on the side are stored as the combination, and the difference total value calculation means is a combination located on the side and a side located on the side. Is calculated as the difference total value. Thus, even when the bidirectional combination exists as the combination, accurate detection is possible.

(9) In the edge direction detection method according to the present invention, an edge detection reliability is calculated for a predetermined edge direction candidate in a rectangular area of M pixels * N pixels, and an edge direction in the rectangular area is calculated based on the edge detection reliability. The edge direction is detected when one pixel is an adjacent pixel adjacent to the central pixel of the rectangular area and the other pixel forms a line segment with the adjacent pixel. At least a plurality of combinations in which the minutes are in parallel relation are stored in each edge direction, and for the designated rectangular area, the absolute value of the difference between the stored combination pixels is summed up for each edge direction. For each edge direction, and each of the combined pixels is based on the pixel value of an adjacent pixel adjacent to the central pixel of the rectangular area. A representative pixel value in the di-direction, a difference between the representative pixel value in the rectangular area from the value of each pixel constituting the specified rectangular area as a representative pixel difference, calculated for each edge direction for each edge direction, wherein a value obtained by multiplying the first coefficient to the difference sum value, calculates a second higher edge detection signal Yoriyukido sum small value obtained by multiplying a coefficient to the central vicinity pixel difference values.

In this way, for each edge direction, the difference between the central neighborhood pixel value and the representative pixel value is calculated as the central neighborhood pixel difference value, and for each edge direction, the difference sum is multiplied by the first coefficient. the value was, the high edge detection first about two sum of values obtained by multiplying the coefficient is less high edge detection signal Yoriyukido made possible the central vicinity pixel difference values.

(11) In the edge direction determination method according to the present invention, edge detection reliability is calculated for a predetermined edge direction candidate in a predetermined area of one screen, and the edge direction in the predetermined area is determined based on the edge detection reliability. In this edge direction determination method, when one pixel is an adjacent pixel adjacent to the central pixel of the rectangular area and the other pixel forms a line segment with the adjacent pixel, the edge direction and the line segment are parallel to each other. At least a plurality of combinations to be related are stored for each edge direction, and the absolute value of the difference of the stored combination pixels is summed for each edge direction with respect to the specified predetermined area. Each edge direction is calculated based on a pixel value of an adjacent pixel adjacent to a central pixel of the predetermined region among the combination pixels. The difference between the representative pixel value in the predetermined direction and the value of each pixel constituting the specified predetermined area is calculated as the representative pixel difference for each edge direction, and the difference total value is An edge direction having a small sum of a value obtained by multiplying a coefficient of 1 and a value obtained by multiplying the center neighboring pixel difference value by a second coefficient is determined as an edge direction to be obtained .

Thus, for each edge direction, the difference between the central neighborhood pixel value and the representative pixel value is calculated as the central neighborhood pixel difference value, and the difference total value is multiplied by the first coefficient, and the center By determining an edge direction having a small sum of values obtained by multiplying the neighboring pixel difference value by the second coefficient as an edge direction to be obtained, edge detection with high reliability can be performed.

( 12 ) The edge direction determination program according to the present invention is a program for causing a computer to function as the following means. When each pixel is an adjacent pixel adjacent to the central pixel of the rectangular area, and the other pixel connects the adjacent pixel and the line segment, a combination in which the edge direction and the line segment are in a parallel relationship, Means for storing at least a plurality of edge directions, means for calculating, for each edge direction, an absolute difference value of the stored combination pixels for a specified rectangular area, and calculating a difference total value for each edge direction; Based on the pixel value of the adjacent pixel adjacent to the central pixel of the rectangular area, the representative pixel value in each edge direction and the representative value in the rectangular area from the value of each pixel constituting the designated rectangular area representative pixel difference the difference between the pixel values, means for computing for each edge direction, and a value obtained by multiplying the first coefficient to the difference sum value, wherein Means for determining the sum is small edge direction of the second value obtained by multiplying the coefficients, as the edge direction determined in central vicinity pixel difference values.

Thus, for each edge direction, the difference between the central neighborhood pixel value and the representative pixel value is calculated as the central neighborhood pixel difference value, and the difference total value is multiplied by the first coefficient, and the center By determining an edge direction having a small sum of values obtained by multiplying the neighboring pixel difference value by the second coefficient as an edge direction to be obtained, edge detection with high reliability can be performed.

  In this specification, the term “edge” means that the above-mentioned features change abruptly when a region having similar features such as the luminance value, density value, color, and pattern of a pixel is defined as one region. , Refers to the boundary between regions. “Edge direction detection” means detecting the direction of this edge. In the embodiment, the “combined pixel storage unit” corresponds to a part that specifies which pixel is used when performing calculations for 12 edge directions in the edge direction detection program 26p.

It is a figure which shows the RGB value of the pixel in a part of screen. It is a figure which shows the window of edge detection in patent document 1. FIG. It is a figure for demonstrating the edge detection method in patent document 1. FIG. It is a figure for demonstrating the edge detection method in patent document 2. FIG. 2 is a functional block diagram of the edge direction detection device 1. FIG. It is a hardware block diagram of the edge direction detection apparatus. It is an edge direction detection flowchart. It is a figure explaining the edge direction candidate employ | adopted with the edge direction detection apparatus 1. FIG. It is a figure explaining the angle of the edge direction candidate employ | adopted with the edge direction detection apparatus 1. FIG. It is a figure which shows the pixel arrangement | positioning of the 4 * 4 window employ | adopted with the edge direction detection apparatus 1. FIG. It is a figure which shows the pixel used for a calculation in the case of the edge directions 1 and 7. FIG. It is a figure which shows the pixel used for a calculation in the case of edge directions 2, 6, 8, and 12. It is a figure which shows the pixel used for a calculation in the case of the edge directions 4 and 10. FIG. It is a figure which shows the pixel used for a calculation in edge direction 3,5,9,11. It is a figure which shows the pixel used for a calculation when respond | corresponding to the error of a horizontal direction and a perpendicular direction. It is a figure for demonstrating a positive / negative group. It is a figure which shows the pixel used for the calculation in a primary group. It is a figure which shows the pixel used for the calculation in a negative group. It is an example of pixel arrangement. It is an example of pixel arrangement. It is an example of pixel arrangement.

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

(1. First embodiment)
(1.1 Function block)
FIG. 5 is a functional block diagram of the edge direction detection device 1 according to the embodiment of the present invention.

  The edge direction detection device 1 calculates edge detection reliability for a predetermined edge direction candidate in a rectangular area of M pixels * N pixels, and detects an edge direction in the rectangular area based on the edge detection reliability. A combined pixel storage means 3, a difference total value calculation means 5, a central neighborhood pixel value calculation means 7, a representative pixel value calculation means 9, a central neighborhood pixel difference value calculation means 11, an edge detection reliability calculation means 13, And determining means 15.

  The combination pixel storage means 3 is such that when one pixel is an adjacent pixel adjacent to the central pixel of the rectangular area and the other pixel forms a line segment with the adjacent pixel, the edge direction and the line segment are in a parallel relationship. Are stored for each of the edge directions.

  The difference total value calculation means 5 calculates, for each edge direction, the difference total value obtained by totaling the absolute values of the difference between the combination pixels in the combination pixel storage means 3 in each edge direction for the designated rectangular area. The center neighborhood pixel value calculation means 7 calculates the representative pixel value in each edge direction based on the pixel value of the adjacent pixel adjacent to the center pixel of the rectangular area among the combination pixels. The representative pixel value calculation means 9 calculates the representative pixel value in the rectangular area from the value of each pixel constituting the designated rectangular area. The center neighborhood pixel difference value calculation means 11 calculates the difference between the center neighborhood pixel value and the representative pixel value as the center neighborhood pixel difference value for each edge direction.

  The edge detection reliability calculation means 13 calculates an edge detection reliability for each edge direction based on the central neighborhood pixel difference value and the difference total value.

The determining unit 15 determines that the edge direction having the high edge detection reliability is the edge direction in the rectangular region.

  Thus, for each edge direction, the difference between the central neighborhood pixel value and the representative pixel value is calculated as the central neighborhood pixel difference value, and for each edge direction, the center neighborhood pixel difference value and the difference total value are calculated. Based on the calculation of the edge detection reliability, edge detection with high reliability can be performed.

(1.2 Hardware configuration)
FIG. 6 shows an example of the hardware configuration of the edge detection apparatus 1. The edge detection apparatus 1 includes a CPU 23, a frame memory 27, a RAM 25, and a flash memory 26. The flash memory 26 stores an edge direction detection program 26p. The edge direction detection program 26p stores combinations of pixels used for calculation and calculation methods for 12 edge directions. Details will be described later.

  The RAM 25 stores calculation results and the like. The frame memory 27 holds image data for one screen.

  The CPU 23 sets a predetermined 4 * 4 pixel window for the image data stored in the frame memory 27 in accordance with the edge direction detection program 26P, reads pixel data (luminance) for the set pixel, and will be described later. The edge detection reliability is obtained, and the result is stored in the RAM 26.

  In the present embodiment, only the edge detection process in one window will be described, but actually, the edge detection process is repeatedly performed by shifting the window by one pixel.

(1.3 Explanation of flowchart)
FIG. 7 shows a flowchart of the program stored in the flash memory 25. When the calculation target window is determined, the CPU 23 calculates a value PX for the window (step S1 in FIG. 7). In the present embodiment, the pixel value PX for validity verification is calculated using an LPF filter in a 4 * 4 pixel window. Specifically, the value PX was determined by equation (4). A pixel calculated using a filter such as Expression (4) indicates a pixel value of an average target position. By using the value PX, it is possible to determine the reliability of the detected edge by the difference from the provisional interpolation pixel generated along each edge.

PX = A * WND4 * A ^T・ ・ ・ Formula (4)?
Here, the matrix A is a parallelization matrix, and the matrix ^AT is a transposed matrix of the matrix A. Next, the CPU 23 performs edge detection reliability reference value SDIF (t) (hereinafter referred to as value SDIF ( t)) (step S3 in FIG. 7).

(1.3.1 Edge direction)
Hereinafter, an edge direction necessary for calculating the value SDIF (t) will be described. In this embodiment, twelve edge directions are adopted as shown in FIGS.

  In FIG. 8A, a virtual pixel G1 is defined at the center of 4 * 4 pixels. In the figure, hatched pixels indicate interpolation pixels. Since there are real pixels C2 to C4 around the virtual pixel G1, interpolation pixels G2 to G5 having a half-pixel relationship are defined inside these pixels. Furthermore, interpolation pixels G12 to G19 having a half-pixel relationship are defined outside the actual pixels C2 to C4.

  By using the interpolation pixel and the real pixel defined in this way, approximately equal 12 directions passing through the virtual pixel G1 are defined. Specifically, as shown in FIG. 8A, a direction 1 connecting the interpolation pixel G2 and the interpolation pixel G4, a direction 2 connecting the real pixel C12 and the real pixel C16, a direction 3 connecting the interpolation pixel G12 and the interpolation pixel G16, and the real pixel The direction 4 connects C2 and the real pixel C4, the direction 5 connects the interpolation pixel G13 and the interpolation pixel G17, and the direction 6 connects the real pixel C13 and the real pixel C17. Similarly, directions 7 to 12 are defined as shown in FIG. 8B. As a result, as shown in FIG. 9, the angles with respect to directions 1 to 12 are defined as 90 degrees, 108.4 degrees, and so on. As shown in FIG. 9, these angular configurations are not completely equal, but are divided into approximately 12 equal parts.

  The 12 directions can be classified into four types of edge types 1 to 4 as processing to be calculated. Edge direction 1 and edge direction 7 are vertical and horizontal directions, and are of edge type 1. This is because the calculation processing pixels can be obtained in the same manner only by being different by 90 degrees. Edge direction 2, edge direction 6, edge direction 8, and edge direction 12 are integer precision directions and are of edge type 2. These can be obtained in the same manner as the arithmetic processing. Edge direction 4 and edge direction 10 are integer precision directions, and are edge type 3. Edge direction 3, edge direction 5, edge direction 9, and edge direction 11 have half-pixel accuracy and are of edge type 4. This is because these can also be obtained in the same manner, although there is a difference between horizontal and vertical.

  FIG. 10 shows a region WND4 that is an example of a region used for the calculation method of the value SDIF (t) in each edge direction. The region WND4 is composed of 4 * 4 real pixels with the virtual pixel G1 as the center. Hereinafter, real pixels are represented by coordinates. That is, the upper left pixel (1, 1), the right pixel (1, 2)... The lower right pixel (4, 4). These pixel values are represented by WND4 (1, 1), (1, 2)..., WND4 (4, 4).

  The value SDIF (t) is obtained by the sum of the pixel difference along the edge direction (SDIF1), the interpolation pixel at the target position using the edge direction, and the difference between the following PX (SDIF2).

(1.3.2 Calculation method for edge types)
Hereinafter, a method of calculating the value SDIF (t) in the edge types 1 to 4 will be described in detail.

(1.3.2.1 Edge type 1)
As shown in FIG. 11A, when the value SDIF (1) in the edge direction 1 is calculated, the pixels (1, 2), (2, 2), (3, 2), (4, 2), (1 , 3), (2, 3), (3, 3), (4, 3) are used.

  The difference DIF1 (t, 2) in the second column from the pixel is obtained by Expression (4-1).

  That is, the difference DIF1 (1, 2) in column 2 is the absolute value of the difference between the value of the pixel (1, 2) and the value of the pixel (2, 2), and the difference DIF1 (2, 2) is the pixel ( 2, 2) and the absolute value of the difference between the pixel (3, 2) value, and the difference DIF1 (3, 2) is the difference between the value of the pixel (3, 2) and the value of the pixel (4, 2). The absolute value of the difference.

  Similarly, the difference DIF2 (t, 3) in the third column is obtained by Expression (4-2).

  The absolute value SDIF1 (1) of these differences is obtained by Expression (4-3). Thus, in the present embodiment, the difference is not added as it is, but the coefficient wt is doubled in the other cases at t = 2. This is for increasing the weight of the combination of the pixels closest to the target pixel, but the weight is not limited to this.

  Next, the interpolated pixel reference value VAL (t) calculated along the direction is calculated by Expression (5-1).

  Specifically, the average value of the pixel (2, 2) and the pixel (3, 2) is used as the interpolation pixel reference value VAL (2), and the pixel (2, 3) is used as the interpolation pixel reference value VAL (3). An average value of the pixels (3, 3) is obtained.

  Next, the difference between the interpolation pixel reference values VAL (2) and PX and the average value SDIF2 of the difference between the interpolation pixel reference value VAL (3) and the value PX are calculated by the equation (6).

  If the edge direction is correct, the difference between the interpolated pixel reference value VAL (t) and the value PX becomes small.

  The sum of the sum SDIF1 and the average value SDIF2 is the value SDIF (1) when the edge direction is 1.

SDIF = SDIF1 + SDIF2 ... Formula (7-0)
In the case of the edge direction 7, compared to the case of the edge direction 1, the pixel position to be used is just 90 degrees rotated. As shown in FIG. 8B, the value SDIF (7) of the edge direction 7 is calculated. Includes pixels (2,1), (2,2), (2,3), (2,4), (3,1), (3,2), (3,3), (3,4) ) To calculate in the same manner.

  The basic concept is the same for the other edge directions, but the pixel used is different and the SDIF (t) calculation method is specifically different.

(1.3.2.2 Edge type 2)
As shown in FIG. 12A, when the value SDIF (2) in the edge direction 2 is calculated, the pixels (1, 1), (4, 2), (1, 2), (4, 3), (1 , 3) and (4, 4) are used.

  The difference DIF (t) from the pixel is obtained by Expression (7-1).

  That is, the difference DIF (1) is the absolute value of the difference between the value of the pixel (1, 1) and the value of the pixel (4, 2), and the difference DIF (2) is the value of the pixel (1, 2). The absolute value of the difference between the values of the pixel (4, 3), and the difference DIF (3) is the absolute value of the difference between the value of the pixel (1, 3) and the value of the pixel (4, 4).

  The sum SDIF1 of the absolute values of these differences is obtained by Expression (7-2). For the edge direction 2 as well, the difference is not added as it is for the same reason as in the edge direction 1, but the coefficient wt is doubled compared to the other cases at t = 2.

  Next, the CPU 23 calculates the interpolated pixel reference value VAL (t) calculated along the direction by the equation (8-1).

  Specifically, an average value of the pixel (1, 2) and the pixel (4, 3) is obtained as the interpolation pixel reference value VAL.

  Next, the CPU 23 calculates the difference between the interpolated pixel reference values VAL and PX as SDIF2 by Expression (8-2).

  The sum of the sum SDIF1 and the average value SDIF2 is the calculated value SDIF (2) when the edge direction is 2.

SDIF = SDIF1 + SDIF2 ... Formula (7-0)
In the case of the edge direction 6, the edge direction 8, and the edge direction 12, as shown in FIGS. 12B to 12D, compared to the case of the edge direction 2, the pixel position to be used is rotated by 90 degrees. Thus, the calculated value SDIF can be obtained.

(1.3.2.3 Edge type 4)
When calculating the value SDIF (4) of the edge direction 4, each of the horizontal direction and the vertical direction can be considered. Therefore, considering the robustness of edge continuity determination, as shown in FIG. 13A, pixels (2,1), (3,2), (2,2), (3,3), (2,3) , (3, 4) and pixels (1, 2), (2, 3), (2, 2), (3, 3), (3, 2), (4, 3) as shown in FIG. ) Is used.

The difference DIF _h (t) in the horizontal direction from the pixel is obtained by Expression (9-1).

That is, the difference DIF _h (1) is the absolute value of the difference between the pixel (2, 1) and the pixel (3, 2), and the difference DIF _h (2) is the value of the pixel (2, 2). The absolute value of the difference between the values of the pixel (3, 3), and the difference DIF _h (3) is the absolute value of the difference between the value of the pixel (2, 3) and the value of the pixel (3, 4).

Similarly, the difference DIF _V (t) in the vertical direction is obtained by Expression (9-2).

The sum SDIF1 _h of the absolute values of the differences in the horizontal direction is obtained by Expression (9-3). For the edge direction 4 as well, the difference is not added as it is for the same reason as in the edge direction 1, but the coefficient wt is doubled compared to the other cases at t = 2.

Similarly, the sum SDIF1 _v of the absolute values of the differences in the vertical direction is obtained by Expression (9-4).

The minimum value of the sum SDIF1 _h and sum SDIF1 _v is SDIF1 (see Equation 10-1).

  Next, the interpolated pixel reference value VAL (t) calculated along the direction is calculated by Expression (10-2).

  Specifically, an average value of the pixel (2, 2) and the pixel (3, 3) is obtained as the interpolation pixel reference value VAL.

  Next, the absolute value SDIF2 of the difference between the interpolation pixel reference values VAL and PX is calculated.

  Similarly, the sum of the sum SDIF1 and the average value SDIF2 is the calculated value SDIF in the case of the edge direction 1.

SDIF = SDIF1 + SDIF2 ... Formula (7-0)
In the case of the edge direction 10, the pixel position to be used is just 90 degrees rotated as compared with the case of the edge direction 4, and as shown in FIG. 13C, the pixels (3, 1), (2, 2), (3,2), (2,3), (3,3), (2,4) and pixels (2,2), (1,3), (3,2) as shown in FIG. 13D , (2, 3), (4, 2), (3, 3) may be used in the same manner.

(1.3.2.4 Edge type 3)
As shown in FIG. 14A, when the value SDIF (3) in the edge direction 3 is calculated, the pixels (2, 1), (4, 2), (1, 1), (3, 2), (2 , 2), (4, 3), (1, 2), (3, 3) (2, 3), (4, 4), (1, 3), (3,4).

  The difference DIF1 (t) is obtained by Expression (11-1).

  That is, the difference DIF1 (1) is the absolute value of the difference between the value of the pixel (1,1) and the value of the pixel (3,2), and the difference DIF1 (2) is the value of the pixel (1,2). The absolute value of the difference between the values of the pixel (3, 3), and the difference DIF1 (3) is the absolute value of the difference between the value of the pixel (1, 3) and the value of the pixel (3, 4).

  Similarly, the difference DIF2 (t) is obtained by Expression (11-2).

  The sum SDIF1 of the absolute values of these differences is obtained by Expression (11-3). For the edge direction 3 as well, the difference is not added as it is for the same reason as in the edge direction 1, but the coefficient wt is doubled in other cases at t = 2. The reason why the weight is set to “1” when t = 2 is because (DIF1 + DIF2).

  Next, the interpolated pixel reference value VAL (t) calculated along the direction is calculated by Expression (12-1).

  Specifically, the average value of the pixel (1, 2) and the pixel (3, 3) is used as the interpolation pixel reference value VAL (1), and the pixel (2, 2) is used as the interpolation pixel reference value VAL (2). An average value of the pixels (4, 3) is obtained.

  Next, the difference between the interpolation pixel reference values VAL (1) and PX and the average value SDIF2 of the difference between the interpolation pixel reference value VAL (2) and the value PX are calculated from the equation (12-2).

  The sum of the sum SDIF1 and the average value SDIF2 is the calculated value SDIF when the edge direction is 1.

SDIF = SDIF1 + SDIF2 ... Formula (7-0)
In the edge directions 5, 9, and 11, the pixel position to be used is just 90 degrees rotated as compared with the edge direction 3, and the value SDIF is calculated as shown in FIGS. be able to.

  In addition, about this edge type, although the pixel combination of 6 was employ | adopted, if it is 3 or more, it will not be limited to this. Alternatively, two may be used if simple calculation is performed. That is, the calculation may be performed from the combination of the pixel (2, 2) and the pixel (4, 3), and the pixel (1, 2) and the pixel (3, 3).

(1.3.3 Determination of edge directions belonging to positive group PG and negative group NG)
A method for determining the positive group PG and the negative group NG using the determination value HOR and the value VER in step S5 in FIG. 7 will be described. First, a method for calculating the determination value HOR and the value VER will be described. The horizontal value HDIF (y) is obtained from Equation (13-1).

  For example, the horizontal value HDIF (1) is the absolute value of the difference between the pixel (1, 2) and the pixel (1, 3) in FIG. Similarly, the maximum value and the minimum value of the values HDIF (1) to HDIF (4) are obtained from the values HDIF (2) to HDIF (4) obtained in the same manner. Similarly, the value VDIF (x) in the vertical direction is obtained from Expression (13-2), and the maximum value and the minimum value of the values VDIF (1) to VDIF (4) are obtained.

  The determination value HOR and the value VER in the horizontal and vertical directions are obtained from the equations (13-3) and (13-4). The value HOR and the value VER are used to determine a case where the value is not horizontal or vertical, as will be described later. The value HOR is the value HOR = if the minimum value of the vertical value VDIF (x) exceeds the maximum value of the horizontal value HDIF (y) by two times or exceeds a predetermined threshold TH1. 1 otherwise, the value HOR = 0. The same applies to the value VER.

  Using the values HOR and VER, the edge direction belonging to the positive group PG, the edge direction belonging to the negative group NG, and the rank in each are determined. Specifically, as shown in FIG. 16, the edge direction angle is divided into a positive edge group PG belonging to an angle of 0 to 90 degrees and an edge direction angle of 90 to 180 degrees, and a negative group NG belonging to the angle group. . That is, the edge directions 8 to 11 belong to the positive edge group PG, and the edge directions 2 to 6 belong to the negative group NG. Further, the edge direction 1 which is the horizontal direction does not belong to any group when the value HOR = 0. Also, the edge direction 7, which is the vertical direction, does not belong to any group if the value VER = 0. In the positive / negative group, the value SDIF for a plurality of edge directions may be the same. In this case, a direction closer to 45 degrees or 135 degrees may be selected. This is because the 45-degree or 135-degree edge determination pixel is close to the pixel of interest and is therefore more reliable than a distant pixel pair.

(1.3.4 Edge group determination)
The edge group determination process in step S7 of FIG. 7 will be described.

  First, as shown in FIG. 17, the differences PDIF (1) to (9) between pixels adjacent in the 45 degree direction are calculated.

  The total value PSUM of the differences PDIF (1) to (9) is calculated. Similarly, the differences NDIF (1) to (9) and the total value NSUM are calculated.

  Next, edge direction determination will be described. In the present embodiment, when the sum of the total value PSUM and the threshold value TH3 is less than or equal to the total value NSUM, “GRP = positive”, and when the sum of the total value NSUM and the threshold value TH3 is less than or equal to the total value PSUM When “GRP = negative” is neither, it is determined that “GRP = 0 (no edge)”.

  In order to determine a positive / negative group, the total value PSUM and the total value NSUM may be compared to determine the group. However, by determining the threshold value TH3 to an appropriate value, it is possible to appropriately determine a case where there is no edge (for example, a checkered pattern).

(1.3.5 Edge direction determination)
The edge direction determination process in step S9 in FIG. 7 will be described. Specifically, when the value GRP is positive, the edge direction having the minimum value in the positive group is selected. If the value GRP is negative, the edge direction having the minimum value in the negative group is selected.

(1.4 About specific examples)
Hereinafter, a specific description will be given based on the flowchart of FIG. 7 using 4 * 4 pixel values. For example, when the rectangular data shown in FIG. 19 is given, the CPU 23 calculates the value PX (step S1 in FIG. 7). In this case, PX = 74.

  Next, the CPU 23 calculates a value SDIF (t) (step S3 in FIG. 7). In this case, SDIF (1) = 181; SDIF (2) = 400; SDIF (3) = 299; SDIF (4) = 131; SDIF (5) = 59; SDIF (6) = 53; SDIF (7) = 114; SDIF (8) = 324; SDIF (9) = 276; SDIF (10) = 229; SDIF (11) = 372; SDIF (12) = 400.

  Next, the CPU 23 calculates values HOR and VER and determines a positive group PG and a negative group NG using the values HOR and VER (step S5 in FIG. 7). In the case of the rectangular data shown in FIG. 19, the value HOR = 0; the value VER = 0;

  Therefore, in this case, the positive group PG is the edge directions “10”, “9”, “11”, “8”, “12”, and the negative group NG is the edge directions “4”, “5”, “3”. ”,“ 6 ”,“ 2 ”.

  Next, the CPU 23 determines which of the positive group PG and the negative group NG is selected based on PSUM and NSUM (step S7 in FIG. 7). In the case of the rectangular data shown in FIG. 19, since the value PSUM = 174; the value NSUM = 414 ;, the value GRP is “negative”.

  Next, the CPU 23 determines the minimum value SDIF (6) = 53 as the edge direction from the determined negative group NG (step S9).

  In this way, the correct edge direction is specified.

(1.5 Significance of mean value SDIF2)
In the present embodiment, the value SDIF is determined for the sum SDIF1 in consideration of the average value SDIF2. Such significance will be described.

  For the 4 * 4 window shown in FIG. 19, the values SDIF1, SDIF2, and SDIF are calculated as follows.

PX = 74;
SDIF1 (4) = 122; SDIF1 (5) = 39; SDIF1 (3) = 289; SDIF1 (6) = 44; SDIF1 (2) = 392;
SDIF2 (4) = 9; SDIF2 (5) = 20; SDIF2 (3) = 10; SDIF2 (6) = 9; SDIF2 (2) = 8;
SDIF (4) = 131; SDIF (5) = 59; SDIF (3) = 299; SDIF (6) = 53; SDIF (2) = 400;
In such a situation, if only the value SDIF1 is selected, the value SDIF1 (5) becomes the minimum value, and the edge direction 5 is selected. In this case, however, the edge direction 5 is incorrect, and the edge direction 6 is correct.

  On the other hand, by using the average value SDIF2, the correct edge direction is selected as described below. In this case, the value PX calculated in step S1 in FIG. 7 is “74”. If the edge direction is 5, the interpolation value along the edge is VAL (1) = (94 + 112) / 2 = 103; VAL (2) = (54 + 72) / 2 = 63. Therefore, in the case of the edge direction 5, SDIF2 = (ABS (103-74) + ABS (63-74)) / 2 = 20.

  On the other hand, if the edge direction is 6, the interpolation value along the edge is VAL = (94 + 72) / 2 = 83, and SDIF2 = ABS (83-74) = 9.

  In this manner, by using the average value SDIF2, detection with fewer errors can be performed.

  Note that the adoption of the average value SDIF2 is not essential.

  Further, in the present embodiment, the value SDIF1 and the value SDIF2 are simply added, but weighting may be performed using a coefficient.

(1.6 Significance of horizontal and vertical judgment)
In the present embodiment, as described above, using the value HOR and the value VER, if these are not “1”, it is determined that the directions are not in the horizontal and vertical directions. Such significance will be described.

  When the value SDIF is calculated for the 4 * 4 window shown in FIG. 21, SDIF (1) = 135; SDIF (2) = 420; SDIF (3) = 259; SDIF (4) = 98; SDIF (5) = 92 SDIF (6) = 68; SDIF (7) = 49; SDIF (8) = 205; SDIF (9) = 179; SDIF (10) = 154; SDIF (11) = 288; SDIF (12) = 420; It becomes.

  Therefore, the edge direction 7 which is SDIF (7) is selected as it is. However, edge direction 6 is correct. Such erroneous selection occurs for the following reason.

  In the horizontal direction, pixels (2, 1) to (2, 4) and pixels (3, 1) to (3,4) are used to calculate the value SDIF (7). In FIG. 21, since all the pixel values of the pixels (3, 1) to (3,4) are the same as “115”, the SDIF becomes small. On the other hand, in the case of the edge direction 6 which is a correct answer, the calculation is performed by a combination of the pixel (1, 1) and the pixel (2, 4), and a combination of the pixel (2, 1) and the pixel (3,4). . In this case, since the difference between the pixel (1, 1) and the pixel (2, 4) is large, the value SDIF (6) in the edge direction 6 is larger than the value SDIF (7) in the edge direction 7. There is a case.

Therefore, in the present embodiment, the case of being horizontal or vertical is extracted with an arithmetic expression different from the value SDIF, and when the condition is not satisfied, the horizontal and vertical edge directions are determined in a later determination step. The candidate was excluded.

  Note that the determination process using the value HOR and the value VER has been described in order to compensate for this, but is not limited to the determination based on the value SDIF.

(1.7 Significance of positive / negative group judgment)
The significance of positive / negative group determination will be described. If the edge direction is specified only by the value SDIF, if the edge width in the window is small, the pixel values on both sides of the edge are almost the same, so there is a risk of erroneous determination.

  For example, when the value SDIF is calculated for the 4 * 4 window shown in FIG. 20, SDIF (1) = 769; SDIF (2) = 124; SDIF (3) = 379; SDIF (4) = 634; SDIF (5) = 379; SDIF (6) = 124; SDIF (7) = 769; SDIF (8) = 634; SDIF (9) = 769; SDIF (10) = 131; SDIF (11) = 769; SDIF (12) = 634. Further, PX = 131; HOR = 0; VER = 0. Therefore, if the positive / negative group judgment is not made, the edge direction 2 or the edge direction 6 is selected.

  On the other hand, by performing the process of step S7 in FIG. 7, the correct edge direction 10 is selected as follows.

  In this case, edge 6 is the smallest SDIF in the priority order among SDIF (4), SDIF (5), SDIF (3), SDIF (6), and SDIF (2) in the negative group. On the other hand, among the SDIF (10), SDIF (9), SDIF (11), SDIF (8), and SDIF (12) of the primary group, the smallest SDIF in the priority order is the edge 10. In this case, the value PSUM = 0; the value NSUM = 1020; thus, the value GRP becomes “positive”. As a result, SDIF (10) is selected.

  Note that the positive / negative group determination process is not essential, and in the present embodiment, the above-described value SDIF has been described in order to compensate for this. However, the determination process is limited to the value SDIF determination. I don't mean.

(2. Other embodiments)
In the above-described embodiment, the case where the rectangular window is 4 * 4 has been described. However, the present invention can be applied to windows of other M * N pixels. In this case, a candidate for the edge direction may be appropriately adopted according to the pixel configuration of the window to be configured. For a window of M * N pixels, it is possible to define (2 (M-1) +2 (N-1)) edge directions. However, it is not necessary to adopt all of them.

  In the present embodiment, the case of using luminance as the pixel value has been described. However, any pixel value may be used as long as the pixel value is used for edge detection. For example, an RGB value may be used.

Further, in the present embodiment, the value SDIF (1) is obtained from the difference between the pixel values. However, when the pixel value is represented by characteristics using a plurality of parameters (for example, RGB values or For the luminance and color components U and V values), a difference vector or a vector distance (for example, a distance index such as ΔE in the case of color) may be the value SDIF (1). The same applies to the value SDIF (2). In the present embodiment, since the case where the rectangular window is 4 * 4 has been described, the virtual pixel G1 is defined in the center of 4 * 4. However, such a virtual pixel may be defined at a position overlapping the actual pixel.

  Further, it may be a window having a predetermined shape that is not a rectangular window.

  In the above embodiment, the pixel value in the vicinity of the center is doubled, but such weighting is arbitrary.

In the above embodiment, the CPU 23 is used to realize the function shown in FIG. 5, and this is realized by software. However, some or all of them may be realized by hardware such as a logic circuit. In addition, you may make it make an operating system (OS) process a part of program.

(3. Disclosure as other inventions)
A part of the apparatus or method disclosed in this specification can be understood as the apparatus or method described below.

(3-1)
In an edge direction detection device that stores a plurality of edge direction candidates for pixels constituting a predetermined area and detects the edge direction based on pixel values located in the edge direction,
The edge direction is a direction parallel to a line segment connecting real pixels in the predetermined region and passing through a rectangular central pixel;
An edge direction detection device characterized by the above.

(3-2)
In an edge direction detection device that stores a plurality of edge direction candidates for pixels constituting a predetermined area and detects the edge direction based on pixel values located in the edge direction,
Obtain the horizontal difference and the vertical difference in the predetermined area,
If the minimum value in the horizontal direction is smaller than the threshold value,
If the minimum value in the vertical direction is greater than the threshold value,
To judge,
An edge direction detection device characterized by the above.

(3-3)
In an edge direction detection device that stores a plurality of edge direction candidates for pixels constituting a predetermined area and detects the edge direction based on pixel values located in the edge direction,
For a combination of pixels adjacent in the positive direction connecting the first quadrant and the third quadrant or the negative direction connecting the second quadrant and the fourth quadrant in the predetermined area, a total difference is obtained for the positive direction and the negative direction. Judging that it is not an edge direction belonging to a large direction,
An edge direction detection device characterized by the above.

Claims (11)

An edge direction detection device that calculates edge detection reliability for a predetermined edge direction candidate in a rectangular area of M pixels * N pixels, and detects an edge direction in the rectangular area based on the edge detection reliability,
When each pixel is an adjacent pixel adjacent to the central pixel of the rectangular area, and the other pixel connects the adjacent pixel and the line segment, a combination in which the edge direction and the line segment are in a parallel relationship, Combination pixel storage means for storing at least a plurality of edge directions,
For the specified rectangular area, the difference total value calculation means for calculating the difference total value obtained by summing up the absolute value of the difference between the combination pixels in the combination pixel storage means for each edge direction for each edge direction,
A central neighborhood pixel value calculating means for calculating a representative pixel value in each edge direction based on a pixel value of an adjacent pixel adjacent to a central pixel of the rectangular region among the combined pixels,
Representative pixel value calculating means for calculating a representative pixel value in the rectangular area from the values of the pixels constituting the designated rectangular area;
For each edge direction, a central neighborhood pixel difference value calculating means for calculating a difference between the center neighborhood pixel value and the representative pixel value as a center neighborhood pixel difference value;
For each edge direction, an edge for calculating a higher edge detection reliability as a sum of a value obtained by multiplying the total difference value by a first coefficient and a value obtained by multiplying the center neighboring pixel difference value by a second coefficient is smaller. Detection reliability calculation means,
Determining means for determining that an edge direction having a high edge detection reliability is an edge direction in the rectangular region;
An edge direction detection device comprising: In the edge direction detection apparatus of Claim 1,
The first and second coefficients are 1;
An edge direction detection device characterized by the above. In the edge direction detection device according to claim 1 or 2,
The edge direction stored in the combination pixel storage means is a direction that is parallel to a line segment that connects real pixels in the rectangular region and passes through a rectangular central pixel;
An edge direction detection device characterized by the above. In the edge direction detection apparatus in any one of Claims 1-3,
The combination of pixels in each edge direction is at least 3, and the weighting coefficient in the center is increased,
An edge direction detection device characterized by the above. An edge direction detection device that calculates edge detection reliability for a predetermined edge direction candidate in a rectangular area of M pixels * N pixels, and detects an edge direction in the rectangular area based on the edge detection reliability,
When each pixel is an adjacent pixel adjacent to the central pixel of the rectangular area, and the other pixel connects the adjacent pixel and the line segment, a combination in which the edge direction and the line segment are in a parallel relationship, Combination pixel storage means for storing at least a plurality of edge directions,
For the specified rectangular area, the difference total value calculation means for calculating the difference total value obtained by summing up the absolute value of the difference between the combination pixels in the combination pixel storage means for each edge direction for each edge direction,
For each edge direction, an edge detection reliability calculation means for calculating a higher edge detection reliability as the difference total value is smaller,
Determining means for determining that an edge direction having a high edge detection reliability is an edge direction in the rectangular region;
With
The combination of pixels in each edge direction stored in the combined pixel storage means is at least 3, and when summing up, the weighting coefficient at the center was increased,
An edge direction detection device characterized by the above. In the edge direction detection apparatus in any one of Claims 1-5,
Find the horizontal difference and vertical difference in the rectangular area,
If the minimum value in the horizontal direction is smaller than the threshold value,
If the minimum value in the vertical direction is greater than the threshold value,
To judge,
An edge direction detection device characterized by the above. In the edge direction detection apparatus in any one of Claims 1-6,
For the combination of pixels adjacent to the positive direction connecting the first quadrant and the third quadrant or the negative direction connecting the second quadrant and the fourth quadrant in the rectangular region, a total difference is obtained for the positive direction and the negative direction. Judging that it is not an edge direction belonging to a large direction,
An edge direction detection device characterized by the above. In the edge direction detection apparatus in any one of Claims 1-7,
The rectangular region is a square;
When the edge direction is parallel to the diagonal of the square, the combination pixel storage unit stores, as the combination, the combination of the combination located on the vertical side of the square and the combination located on the horizontal side. And
The difference total value calculation means calculates the sum of both the combination located on the vertical side and the combination located on the horizontal side as the difference total value;
An edge direction detection device characterized by the above. An edge direction detection method for calculating edge detection reliability for a predetermined edge direction candidate in a rectangular area of M pixels * N pixels, and detecting an edge direction in the rectangular area based on the edge detection reliability,
When each pixel is an adjacent pixel adjacent to the central pixel of the rectangular area, and the other pixel connects the adjacent pixel and the line segment, a combination in which the edge direction and the line segment are in a parallel relationship, Store at least multiple edge directions,
For the specified rectangular area, the absolute value of the difference between the stored combination pixels is calculated for each edge direction, and a total difference value obtained by summing up the values for each edge direction is calculated.
Based on the pixel value of an adjacent pixel adjacent to the central pixel of the rectangular area among the combined pixels, the representative pixel value in each edge direction and the value of each pixel constituting the designated rectangular area The difference from the representative pixel value in the region is calculated as the representative pixel difference for each edge direction,
For each edge direction, the difference between the central neighborhood pixel value and the representative pixel value is calculated as the central neighborhood pixel difference value,
For each edge direction, calculating a higher edge detection reliability as the sum of a value obtained by multiplying the difference total value by a first coefficient and a value obtained by multiplying the center neighboring pixel difference value by a second coefficient is smaller;
An edge detection reliability calculation method comprising: An edge direction determination method for calculating edge detection reliability for a predetermined edge direction candidate in a predetermined area of one screen and determining an edge direction in the predetermined area based on the edge detection reliability,
When each pixel is an adjacent pixel adjacent to the central pixel of the rectangular area, and the other pixel connects the adjacent pixel and the line segment, a combination in which the edge direction and the line segment are in a parallel relationship, Store at least multiple edge directions,
For the specified area, the absolute value of the difference between the stored combination pixels is calculated for each edge direction, and the total difference value calculated for each edge direction is calculated for each edge direction.
Based on the pixel value of an adjacent pixel adjacent to the central pixel of the predetermined region among the combination pixels, the predetermined pixel value is determined from the representative pixel value in each edge direction and the value of each pixel constituting the specified predetermined region. The difference from the representative pixel value in the region is calculated as the representative pixel difference for each edge direction,
For each edge direction, the difference between the central neighborhood pixel value and the representative pixel value is calculated as the central neighborhood pixel difference value,
Determining an edge direction having a small sum of a value obtained by multiplying the difference total value by a first coefficient and a value obtained by multiplying the center neighboring pixel difference value by a second coefficient as an edge direction to be obtained;
Edge direction determination method comprising: An edge direction determination program for causing a computer to function as the following means.
When one pixel is an adjacent pixel adjacent to the central pixel of the rectangular area of M pixels * N pixels, and the other pixel forms a line segment with the adjacent pixel, the edge direction and the line segment are in a parallel relationship. Means for storing at least a plurality of combinations for each edge direction;
Means for calculating, for each edge direction, a difference total value obtained by summing up the absolute values of the differences of the stored combination pixels with respect to each edge direction for the designated rectangular area;
Based on the pixel value of an adjacent pixel adjacent to the central pixel of the rectangular area among the combined pixels, the representative pixel value in each edge direction and the value of each pixel constituting the designated rectangular area Means for calculating the difference between the representative pixel value in the region as the representative pixel difference for each edge direction;
Means for calculating a difference between a central neighborhood pixel value and the representative pixel value as a central neighborhood pixel difference value for each edge direction;
Means for determining an edge direction having a small sum of a value obtained by multiplying the difference total value by a first coefficient and a value obtained by multiplying the center neighboring pixel difference value by a second coefficient as an edge direction to be obtained.