JP5940403B2 - Edge direction discriminator - Google Patents

Description

  The present invention relates to an edge direction discriminating apparatus that discriminates the edge direction of each pixel in an image.

By determining the edge direction of each pixel in the image, it is possible to simplify subsequent image processing such as noise removal while retaining the structural attributes of the image. Improvement is desired.
As a technique related to edge direction discrimination, for example, Japanese Patent Application Laid-Open No. 2008-293425 (Patent Document 1) discloses that a plurality of band image signals having different frequency bands are obtained by performing multi-resolution conversion on an input image signal. A first image signal that includes information on a frequency band of the band image signal and information on a frequency lower than the frequency band, and information on a frequency lower than the frequency band of the band image signal. There is disclosed a noise removing device that determines the direction of an edge component of a band image signal using two image signals and removes noise of the band image signal in accordance with the direction of the edge component.

JP 2008-293425 A

However, in the noise removal apparatus described in Patent Document 1 described above, noise removal processing is performed based on directional filter processing, so that there is an advantage of high edge storage capability, but it exists in a flat portion or the like. There is a tendency to capture and save minute edges as edges, and pattern-like noise may remain in the output image.
Japanese Patent Application Laid-Open No. 2004-228620 proposes to hold the edge by performing a coring process after the direction determination process in order to remove pattern noise. However, when the amount of noise is large and pattern-like noise is conspicuous, it is necessary to increase the intensity of the coring process in order to remove noise, which causes a problem that the image quality deteriorates.

  The present invention has been made in view of the above-described circumstances, and an object thereof is to determine the direction of the edge of a pixel of interest with higher accuracy regardless of the amount of noise.

In order to achieve the above object, the present invention provides the following means.
The present invention has a first direction discriminating means for discriminating the direction of an edge in a target pixel and outputting a first discrimination result, and has a low-pass effect higher than that of the first direction discriminating means. A second direction discriminating unit that discriminates a direction and outputs a second discrimination result; a flatness calculating unit that calculates the flatness of the pixel of interest; and a flatness calculated by the flatness calculating unit. As a combination ratio of the first determination result and the second determination result, the higher the flatness, the higher the ratio of the first determination result, and the lower the flatness, the higher the ratio of the second determination result. Combining ratio calculating means for calculating the combined ratio, and combining means for combining the first determination result and the second determination result based on the combining ratio calculated by the combining ratio calculating means, Paired with the result of synthesis by synthesis means An edge direction discriminating apparatus for discriminating a corresponding direction as an edge direction of a target pixel is provided.

  According to the present invention, the direction of the edge of the pixel of interest is discriminated by the two direction discriminating units of the first direction discriminating unit and the second direction discriminating unit having a low pass effect higher than that of the first direction discriminating unit. And the second discrimination result are output. Here, the first direction discriminating means has a drawback that the low-pass effect is low as compared with the second direction discriminating means, that is, the noise removal effect is small, and thus the edge holding force is weak. On the other hand, the second direction discriminating unit has a high low-pass effect as compared with the first direction discriminating unit, that is, the noise removing effect is high. While the holding force is strong, there is a risk that minute noise such as a flat portion may be identified as an edge. Therefore, the flatness calculation means calculates the flatness of the target pixel, combines the first determination result and the second determination result based on the combination ratio according to the flatness, and determines the direction corresponding to the combination result. The edge direction of the pixel of interest is determined. Therefore, the direction of the edge of the target pixel can be determined with higher accuracy regardless of the amount of noise.

In the above-described invention, the first determination result and the second determination result are vector values, and the combining unit weights and adds the vector values based on the combination ratio calculated by the combination ratio calculation unit. Accordingly, it is preferable to combine the first determination result and the second determination result.
In this way, since the calculation at the time of combining the first discrimination result and the second discrimination result can be simplified, the target pixel can be easily and highly accurately irrespective of the amount of noise. The edge direction can be determined with higher accuracy.

In the above-described invention, the first discrimination result and the second discrimination result are numerical values obtained by discretizing an angle or an angle, and the synthesizing unit calculates the synthetic ratio calculating unit to a numerical value obtained by discretizing the angle or the angle. It is preferable that the first discrimination result and the second discrimination result are synthesized by multiplying and adding the synthesis ratio calculated by the above.
By doing so, the calculation at the time of synthesizing the first discrimination result and the second discrimination result can be simplified, so that the pixel of interest can be easily and highly accurately irrespective of the amount of noise. The direction of the edge can be determined with higher accuracy.

  According to the present invention, there is an effect that the direction of the edge of the pixel of interest is determined with higher accuracy regardless of the amount of noise.

It is a block diagram showing a schematic structure of an edge direction discriminating device concerning one embodiment of the present invention. FIG. 4 is a conceptual diagram illustrating an example of a filter set by a first direction determining unit in order to determine an edge of a target pixel in an edge direction determining device according to an embodiment of the present invention. FIG. 4 is a conceptual diagram illustrating an example of a filter set by a second direction determining unit in order to determine an edge of a target pixel in an edge direction determining device according to an embodiment of the present invention. In the edge direction discriminating device concerning one embodiment of the present invention, it is a graph which shows the relation between flatness and composition rate when one threshold is set up. In the edge direction discriminating device concerning one embodiment of the present invention, it is a graph which shows the relation between flatness and composition rate when two thresholds are set up. 5 is a flowchart for determining an edge direction of a pixel of interest in an edge direction determination device according to an embodiment of the present invention. FIG. 6 is a conceptual diagram illustrating an example of a filter that is set to determine the edge of a target pixel in an edge direction determination device according to a modified example of an embodiment of the present invention, and (A) illustrates a first direction determination unit; (B) shows an example of the filter of the second direction discriminating unit. 1 is a block diagram illustrating a schematic configuration of an image processing apparatus to which an edge direction determination device according to an embodiment of the present invention is applied.

Hereinafter, an edge direction discriminating apparatus 1 according to an embodiment of the present invention will be described with reference to the drawings.
As shown in FIG. 1, the edge direction determination device 1 includes a first direction determination unit 10, a second direction determination unit 11 that determines the direction of an edge in a target pixel of an input image, and a flatness that calculates the flatness of the target pixel. The calculation unit 12, the synthesis ratio calculation unit 13 that calculates the synthesis ratio of the output results of the direction determination units 10 and 11 according to the flatness, and the output result of the direction determination units 10 and 11 based on the synthesis ratio. A synthesis unit 14 is provided.

  The first direction discriminating unit 10 discriminates the direction of the edge of each pixel included in the input image and outputs the output result to the synthesizing unit 14 described later. The determination of the edge direction by the first direction determination unit 10 is performed as follows, for example.

  First, the first direction discriminating unit 10 extracts a predetermined range including a predetermined number of pixels around a target pixel that is a pixel whose edge direction is to be discriminated. In other words, the first direction discriminating unit 10 uses a filter in a predetermined range including a predetermined number of pixels centered on the pixel of interest in order to determine the direction of the edge. In the present embodiment, the first direction determination unit 10 will be described as using a 5 × 5 pixel filter centered on the target pixel. FIG. 2 illustrates a square filter including a total of 25 pixels (X00 to X44) of 5 pixels × 5 pixels centered on the pixel of interest X22. As shown in FIG. 2, it is assumed that the filter has eight directions e0 to e7 that start radially from the target pixel X22.

  The first direction discriminating unit 10 calculates the difference between the target pixel X22 and each comparison pixel for each of the directions e0 to e7 according to the following formulas 1-0 to 1-7, and sums the absolute values of the differences for each direction. Is output as an evaluation value in the direction. That is, the following calculation is performed for each direction, and the calculation result is used as an evaluation value for the direction. Then, the first direction determination unit 10 outputs the direction indicating the minimum value among the evaluation values for each direction to the synthesis unit 14 as the first determination result.

  Similar to the first direction discriminator 10, the second direction discriminator 11 discriminates the edge direction of each pixel included in the input image and outputs the output result to the synthesizing unit 14 described later. Here, the second direction determination unit 11 has a low-pass effect, that is, a noise removal effect higher than that of the first direction determination unit 10, and the second direction determination unit 11 determines the edge direction. For example, it is performed as follows.

  The second direction discriminating unit 11 extracts a predetermined range including a predetermined number of pixels around a target pixel which is a pixel whose edge direction is to be discriminated. At this time, since the second direction determination unit 11 has a higher low-pass effect than the first direction determination unit 10, for example, the filter size is larger than the filter in the first direction determination unit. In the present embodiment, the second direction determination unit 11 will be described as using a 7 × 7 pixel filter centered on the target pixel. FIG. 3 exemplifies a square filter including 49 pixels (Y00 to Y66) of 7 pixels × 7 pixels centered on the target pixel Y33. As shown in FIG. 3, it is assumed that the filter has eight directions e0 to e7 that start radially from the target pixel Y33.

  The second direction discriminating unit 11 calculates the difference between the target pixel Y33 and each comparison pixel for each of the directions e0 to e7 according to the following formulas 2-0 to 2-7, and sums the absolute values of the differences for each direction. Is output as an evaluation value in the direction. That is, the following calculation is performed for each direction, and the calculation result is used as an evaluation value for the direction. Then, the second direction determination unit 11 outputs the direction indicating the minimum value among the evaluation values for each direction as the second determination result to the synthesis unit 14.

  In addition, each above-mentioned direction e0-e7 can be represented by the label which labeled these with the angle, the direction vector, or these, for example. In the following, an example is shown in which labels are assigned to the directions e0 to e7, and these labels are associated with vector values.

  The flatness calculation unit 12 calculates the flatness of the target pixel. Here, the flatness is calculated for each pixel of the input image and indicates the degree to which the target pixel is flat. Therefore, when the flatness is high, it can be determined that the target pixel is high in flatness, that is, there is no edge. On the other hand, when the flatness is low, the degree of flatness of the target pixel is low, and it can be determined that there is a pattern such as a fine texture or an edge.

  The flatness calculation unit 12 calculates a dispersion value indicating variation in evaluation values in the directions e0 to e7 obtained by the second direction determination unit, and calculates this dispersion value as a value indicating flatness. Accordingly, the flatness calculation unit 12 determines that the flatness is high when the calculated dispersion value is high, and determines that the flatness is low when the dispersion value is low.

  The flatness calculation is not limited to the above-described method. For example, for a plurality of blocks near the target pixel, the maximum value of the pixel value difference between the target pixel and the pixel included in each block is calculated. The flatness can also be determined based on the minimum value among the maximum values for the block.

  The combination ratio calculation unit 13 combines the first determination result and the second determination result according to the flatness calculated by the flatness calculation unit 12, that is, according to the variance value in the present embodiment. Is calculated. As the synthesis ratio, the higher the flatness, the higher the ratio of the first discrimination result, and the lower the flatness, the higher the ratio of the second discrimination result. The calculated composition ratio is output to the composition unit 14.

  In calculating the composition ratio, one or a plurality of threshold values can be determined in advance. An example in the case of one threshold is shown in FIG. 4, and an example in the case of two thresholds is shown in FIG. When one threshold Th1 is set, for example, as shown in FIG. 4, the ratio of the first determination result is the minimum and the ratio of the second determination result is the maximum until the flatness is Th1. A constant composition ratio is calculated, and for Th1 or more, a composition ratio is calculated such that the ratio of the first determination result increases as the flatness increases.

  When two threshold values, Th1 and Th2, are set, for example, as shown in FIG. 5, the ratio of the first determination result is the minimum and the ratio of the second determination result until the flatness is Th1. Is calculated so that the ratio of the first determination result increases as the flatness increases between Th1 and Th2. Then, when the flatness is equal to or greater than Th2, a fixed composition ratio is calculated in which the ratio of the first determination result is the maximum and the ratio of the second determination result is the minimum.

  The combining unit 14 combines the first determination result and the second determination result based on the combination ratio calculated by the combination ratio calculation unit 13. That is, each discrimination result is multiplied by a combination rate for each discrimination result based on the combination ratio calculated by the synthesis ratio calculation unit 13, and these are added together and output as a synthesis result.

When each determination result is a vector value, each x component and y component are weighted and calculated according to the following equations (3) and (4).
Regarding the first discrimination result, the x component is DX1, the y component discrimination result is DY1, the x component synthesis rate is WX1, and the y component synthesis rate is WX1.
Regarding the second discrimination result, the x component is DX2, the y component discrimination result is DY2, the x component synthesis rate is WX2, and the y component synthesis rate is WX2.

X component synthesis result = DX1 × WX1 + DX2 × WX2 (3)
Y component synthesis result = DY1 × WY1 + DY2 × WY2 (4)

  The edge direction discriminating apparatus 1 finally discriminates the direction corresponding to the synthesis result by the synthesis unit 14 as the edge direction of the target pixel.

  Next, the operation of the edge direction discriminating apparatus 1 configured as described above will be described with reference to the flowchart of FIG. In the following description, FIG. 2 and FIG. 3 will be referred to, assuming that the first direction discriminating unit uses a 5 pixel × 5 pixel filter, and the second direction discriminating unit uses a 7 pixel × 7 pixel filter. .

  In order to discriminate the direction of the edge in the image by the edge direction discriminating apparatus 1 according to the present embodiment, the pixel of interest for which the direction of the edge is to be discriminated is specified. A range of 5 pixels by 5 pixels is defined, and a difference between the pixel of interest X22 and each comparison pixel is calculated for each direction of e0 to e7 with respect to this range, and the sum of absolute values of the differences is calculated for each direction. A direction evaluation value is calculated. And the direction which shows the minimum value among each evaluation value is output to the synthetic | combination part 14 as a 1st discrimination | determination result. In the next step S12, a range of 7 pixels × 7 pixels centered on the pixel of interest Y33 is defined by the second direction discriminating unit, and the pixel of interest Y33 and each comparison pixel are compared with this range in each of the directions e0 to e7. The difference is calculated, and the absolute value sum of the differences is calculated for each direction, and the evaluation value for the direction is calculated. Then, the direction indicating the minimum value among the evaluation values is output to the synthesis unit 14 as the second discrimination result.

  In the next step S13, the flatness of the target pixel is calculated. As described above, the flatness calculation unit calculates a dispersion value indicating the variation in the evaluation value in each direction e0 to e7 obtained by the second direction determination unit, and calculates this dispersion value as a value indicating the flatness. And output to the composite ratio calculation unit 13.

  In the next step S <b> 14, the composition ratio calculation unit 13 calculates a composition ratio based on the flatness (dispersion value) and outputs the composition ratio to the composition unit 14. In step S <b> 15, the combining unit 14 combines the first determination result and the second determination result based on the combination ratio calculated by the combination ratio calculation unit 13, and thereby the edge direction determination device 1 serves as the combination result. Is finally determined as the edge direction of the target pixel.

  As described above, the direction of the edge of the target pixel is once determined by the two direction determination units having different characteristics such as the noise removal effect or the direction determination accuracy, and 2 based on the combination ratio corresponding to the flatness of the target pixel. By combining the two determination results and determining the direction corresponding to the combination result as the edge direction of the target pixel, the direction of the edge of the target pixel can be determined with higher accuracy regardless of the amount of noise.

(Modification)
In the above-described embodiment, the first direction discriminating unit and the second direction discriminating unit are exemplified as those having different filter sizes. However, the present invention is not limited to this. For example, filters having the same filter size and different coefficients Can also be used.

  In this case, in order to make the low-pass effect different, the low-pass effect is reduced by increasing the weight of the pixel that is close to the target pixel compared to the weight of the far pixel, and the pixel that is close to the target pixel is also far. By setting the pixels to the same weight, the low pass effect is increased. The direction can be 0 to 360 degrees, and the direction discrimination result here is the gradient direction of the edge.

FIG. 7A shows a 3 pixel × 3 pixel filter having a relatively low low-pass effect used in the first direction discriminating unit, and FIG. 7B shows 3 pixels × a relatively high low-pass effect used in the second direction discriminating unit. A three pixel filter was illustrated.
The first direction discriminating unit calculates the value Tx1 of the value obtained by applying the vertical filter in FIG. 7A and the value Yx1 of the value obtained by applying the horizontal filter. The direction is calculated by substituting those values into the following equation (5-1).

  Similarly, the second direction discrimination processing unit calculates the value of the output Tx2 obtained by applying the vertical filter in FIG. 7B and the value of the output Yx2 obtained by applying the horizontal filter. The direction is calculated by substituting these values into the following equation (5-2).

上記計算式より得られた、direction1を第１の判別結果とし、direction2を第２の判別結果として出力し、平坦度算出、合成比率算出等以後の演算を行い、最終的にエッジ方向を判別する。

The direction 1 obtained from the above formula is output as the first discrimination result, the direction 2 is output as the second discrimination result, and the subsequent calculations such as flatness calculation and composition ratio calculation are performed, and the edge direction is finally discriminated. .

  By applying the edge direction discriminating apparatus 1 described above to an image processing apparatus, it is possible to perform noise reduction processing on the image in which the edge direction of each pixel is discriminated using information on the discriminated direction. In this case, as illustrated in FIG. 8, for example, the image processing device 2 including the image input unit 20, the memory 21, the edge direction determination device 1, and the image processing unit 22 can be provided. The image processing unit 22 includes a smoothing unit 25 along the direction, a smoothing unit 26 of a flat part, and an image composition unit 27 that combines the original image output from the memory 21 and the smoothed image.

  In the image processing unit 22, based on the direction determination result and the flatness result obtained by the edge direction determination device 1, if the image processing unit 22 is an edge and has a direction, the smoothing unit 25 performs smoothing along the direction. In the case of flatness that is not an edge, smoothing using a predetermined block is performed by the smoothing unit 26 and finally synthesized with the original image output from the memory 21.

  In the smoothing process, the average value of a plurality of surrounding pixels along a determined direction of a certain pixel of interest is set to the value of the pixel of interest, and if there is no direction, the average value of a plurality of pixels around a certain pixel of interest is calculated. It may be smoothed by using it. Thereby, the noise on the edge can be dropped without dulling the edge.

DESCRIPTION OF SYMBOLS 1 Edge direction discrimination apparatus 2 Image processing apparatus 10 1st direction discrimination | determination part 11 2nd direction discrimination | determination part 12 Flatness calculation part 13 Synthesis | combination ratio calculation part 14 Synthesis | combination part

Claims (3)

First direction discrimination means for discriminating the direction of the edge of the pixel of interest and outputting a first discrimination result;
A second direction determining unit that has a low-pass effect higher than that of the first direction determining unit, determines a direction of an edge in the target pixel, and outputs a second determination result;
Flatness calculating means for calculating the flatness of the target pixel;
As a combination ratio of the first determination result and the second determination result according to the flatness calculated by the flatness calculation means, the higher the flatness, the higher the ratio of the first determination result, A composition ratio calculating means for calculating a composition ratio in which the ratio of the second determination result is increased as the flatness is lower;
Combining means for combining the first determination result and the second determination result based on the combination ratio calculated by the combination ratio calculation means;
An edge direction discriminating apparatus that discriminates a direction corresponding to a synthesis result by the synthesizing unit as an edge direction of a target pixel. The first discrimination result and the second discrimination result are vector values;
2. The combination according to claim 1, wherein the combining unit combines the first determination result and the second determination result by weighted addition of vector values based on the combination ratio calculated by the combination ratio calculation unit. Edge direction discriminator. The first discrimination result and the second discrimination result are angles or numerical values obtained by discretizing angles,
The combining means combines the first determination result and the second determination result by multiplying the angle or a numerical value obtained by discretizing the angle by the multiplication ratio calculated by the combining ratio calculation means and adding the result. The edge direction discriminating apparatus according to claim 1.

Images