TWI807627B - Image edge detecting method and image edge processing device - Google Patents

Abstract

An image edge detecting method for processing an image including a plurality of pixels includes respectively performing a convolution calculation with a first direction gradient operator and a second direction gradient operator for the image to obtain a first direction gradient information and a second direction gradient information, wherein the first direction is perpendicular to the second direction; determining gradient statistics of a neighboring area of a target pixel of the image according to the first direction gradient information and the second direction gradient information; and determining an edge significance of the target pixel according to the gradient statistics.

Description

Image edge detection method and image edge detection device

The present invention refers to an image edge detection method and image edge detection device, especially an image edge detection method and image edge detection device applied to image noise reduction and image sharpening.

Existing image processing techniques include judging whether an edge in an image has a direction by an edge detection method, for example, an edge detection method can judge whether an edge has a direction according to an edge gradient. However, when the edge detection method encounters weak edges or strong textures, the judgment may be inaccurate, which will affect the edge information required for image sharpening and image noise reduction, resulting in blurred edges or obvious edge noise.

Specifically, the traditional edge detection technology calculates gradients in different directions of an image according to a Sobel operator, and regards regions with larger gradients as having directional edges. However, when the gradient of the directional texture is not large enough, the above method may identify the texture as an edge and cause misjudgment, or, when the directional edge is weak, the above method is also difficult to detect the directional edge. Therefore, there is a need for improvement in the prior art.

In view of this, the main purpose of the present invention is to provide a kind of image edge detection method and An image edge detection device to effectively detect directional edges in an image.

The embodiment of the present invention discloses an image edge detection method for processing an image comprising a plurality of pixels. The method includes performing a convolution operation on the image with a gradient operator in a first direction and a gradient operator in a second direction to obtain gradient data in a first direction and gradient data in a second direction, wherein the first direction is perpendicular to the second direction; according to the gradient data in the first direction and the gradient data in the second direction, gradient statistics are calculated in an adjacent region of a target pixel of the image to obtain a gradient statistic; The edge saliency of one of the target pixels.

The embodiment of the present invention further discloses an image edge detection device for processing an image comprising a plurality of pixels, which includes a convolution circuit, including a first convolution circuit unit and a second convolution circuit unit, which are used to respectively perform a convolution operation on the image with a gradient operator in the first direction and a gradient operator in the second direction to obtain gradient data in the first direction and gradient data in the second direction, wherein the first direction is perpendicular to the second direction; Gradient statistics are calculated in a neighboring region of the pixel to obtain a gradient statistic; and a judging circuit is used to determine an edge saliency corresponding to the target pixel according to the gradient statistic.

10: Image edge detection method

102-112: Steps

20,50: Image edge detection device

202,502: convolutional circuits

2022: The first convolutional circuit unit

2024: The second convolutional circuit unit

204,504: product circuits

206,506: Statistical circuits

208,508: judgment circuit

210: processing circuit

5022,5024: convolutional circuit unit

f(Je): edge saliency function

I: Image

Ixx (i,j), Iyy (i,j), Ixy (i,j): gradient statistics

Je: edge salience

Y: channel

UV: channel

YUV: Image

FIG. 1 is a schematic diagram of an image edge detection method according to an embodiment of the present invention.

FIG. 2 is a schematic diagram of an image edge detection device according to an embodiment of the present invention.

FIG. 3 is a schematic diagram of a mapping curve of a mapping function according to an embodiment of the present invention.

FIG. 4 is a schematic diagram of an application scenario of an embodiment of the present invention.

Fig. 5 is a schematic diagram of an image edge detection device according to another embodiment of the present invention.

Please refer to FIG. 1 and FIG. 2 at the same time. FIG. 1 is a schematic diagram of an image edge detection method 10 according to an embodiment of the present invention, and FIG. 2 is a schematic diagram of an image edge detection device 20 according to an embodiment of the present invention. The image edge detection device 20 can be used to execute the image edge detection method 10 to process an image I including a plurality of pixels. The image edge detection device 20 includes a convolution circuit 202 , a product circuit 204 , a statistics circuit 206 , a judgment circuit 208 and a processing circuit 210 .

In step 104, the image edge detection method 10 uses a gradient operator (Operator) in the first direction and a gradient operator in the second direction to perform a convolution operation on the image I respectively to obtain a gradient data Gx in the first direction and a gradient data Gy in the second direction, wherein the first direction is perpendicular to the second direction.

Specifically, the first convolution circuit unit 2022 and the second convolution circuit unit 2024 of the convolution circuit 202 can respectively perform convolution operations on all pixels of the image I using gradient operators in the first direction and gradient operators in the second direction. In one embodiment, the first convolution circuit unit 2022 can perform convolution operation on the image I with the gradient operator Kx in the x direction to obtain the gradient data Gx in the first direction, and the second convolution circuit unit 2024 can perform convolution operation on the image I with the gradient operator Ky in the y direction to obtain the gradient data Gy in the second direction, and the above-mentioned x direction (ie, the first direction) and the y direction (ie, the second direction) are perpendicular to each other.

For example, when the coordinates of the target pixel A in the image I are (i, j), the image edge detection device 20 can obtain the first direction gradient data Gx and the second direction gradient data Gy of the target pixel A according to the step 104 of the image edge detection method 10, as shown in the following formulas (1)-(4): Kx =[-1 0 1]...(1)

Ky = [-1 0 1] ^T ... (2)

Among them, Kx is the gradient operator in the x direction, Ky is the gradient operator in the y direction, Gx(i,j) represents the result of convolution between the target pixel A and the gradient operator Kx in the x direction, and Gy(i,j) represents the result of convolution between the target pixel A and the gradient operator Ky in the y direction.

Since the existing image edge detection technology only uses the gradient information of a single pixel when determining the edge directionality of the image, it cannot effectively identify the texture or weak edge in the image, which further affects the accuracy of image edge identification. Therefore, the image edge detection method 10 of the embodiment of the present invention further uses the pixel gradient information of the adjacent area of the target pixel A for identifying the image edge, wherein the adjacent area may include at least one adjacent pixel of the target pixel A. In an embodiment, the adjacent area may be adjacent pixels of 5*5 or 7*7 pixel units adjacent to the target pixel A.

In step 106 , the image edge detection method 10 calculates gradient statistics in the vicinity of the target pixel A in the image I according to the first directional gradient data Gx and the second directional gradient data Gy to obtain a gradient statistic.

In practice, the product circuit 204 can be used to calculate the square Gxx=GxGx of the first direction gradient data Gx, the square Gyy=GyGy of the second direction gradient data Gy, and the first direction gradient data Gx according to the first direction gradient data Gx and the second direction gradient data Gy of the target pixel A. The product Gxy with the second direction gradient data Gy=GxGy.

Then, the statistical circuit 206 can calculate the gradient statistics in the vicinity of the target pixel A according to a weighting coefficient, the square Gxx of the first direction gradient data Gx, the square Gyy of the second direction gradient data Gy, and the product Gxy of the first direction gradient data Gx and the second direction gradient data Gy, wherein the weighting coefficient can be a Gaussian weighted or a mean weighted.

When the weighting coefficient w is a Gaussian weighting coefficient, it can be shown as formula (5):

When the weighting coefficient w is the mean coefficient, it can be shown in formula (6): w(m,n)=1...(6)

Wherein, σ is the standard deviation of a normal distribution, and the values of m and n are the range of the adjacent area of the target pixel A. For example, when the adjacent area is a 5*5 pixel unit, m and n are integers between -2 and 2 (ie -2, -1, 0, 1, 2); when the adjacent area is a 7*7 pixel unit, then m and n are integers between -3 and 3 (ie -3, -2, -1, 0, 1, 2, 3).

Therefore, the gradient statistics Ixx (i, j), Iyy (i, j), Ixy (i, j) can be shown as formula (7), (8), (9):

It should be noted that the steps 104 and 106 of the image edge detection method 10 only describe the calculation of the gradient statistics Ixx (i,j), Iyy (i,j), and Ixy (i,j) of adjacent pixels of the target pixel A for a single target pixel A. In fact, the image edge detection device 20 of the embodiment of the present invention calculates the corresponding gradient statistics Ixx (i, j), Iyy (i, j), and Ixy (i, j) for all pixels in the image I according to the image edge detection method 10.

In this case, the image edge detection method 10 of the embodiment of the present invention uses the gradient information of the target pixel A and the pixels in the adjacent area adjacent to the target pixel A as the basis for determining the edge directionality of the image, so as to improve the accuracy of image edge directionality judgment.

In step 108 , the image edge detection method 10 determines an edge saliency Je corresponding to the target pixel A according to the gradient statistics Ixx (i,j), Iyy (i,j), and Ixy (i,j).

Specifically, the edge saliency degree Je may be determined according to a covariance matrix (covariance matrix), one of the second-order statistics of the gradient statistics Ixx (i,j), Iyy (i,j), and Ixy (i,j). For example, a covariance matrix Cxy of the gradient statistics Ixx (i, j), Iyy (i, j), Ixy (i, j) can be shown as formula (10):

When the difference between the eigenvalue λ ₁ of the main direction and the eigenvalue λ ₂ of the primary direction of the covariance matrix Cxy is larger, it means that the edge saliency of the target pixel A is greater, wherein the main direction and the secondary direction are orthogonal to each other. Since the gradient statistics Ixx (i, j), Iyy (i, j), and Ixy (i, j) are obtained according to the gradient information of the target pixel A and the pixels in its adjacent area, when λ ₁ ≫ λ ₂ , the corresponding adjacent area may be an edge area.

In one embodiment, the image edge detection method 10 of the embodiment of the present invention can set the edge salient degree Je as:

When the difference between λ ₁ and λ ₂ is larger, the edge salience is greater; when λ ₁ and λ ₂ are the same, the edge salience is Je=2. Besides, formula (11) can also be deduced into equivalent formula (12):

In practice, the judgment circuit 208 can be used to judge the edge salience degree Je, and the judgment circuit 208 can calculate the edge salience degree Je by using formula (11) or formula (12).

In step 110 , the image edge detection method 10 normalizes the edge saliency Je corresponding to the image I, and determines the edge data of the image I in a mapping manner.

Since the range of the edge saliency Je is [2,∞), it needs to be normalized to be between 0 and 1 in actual use, and be properly mapped to meet the requirements of the application scenario. The image edge detection method 10 of the embodiment of the present invention normalizes the edge saliency Je corresponding to the image (Normalized), and determines the edge data Jec of the image by a mapping method, which can be expressed as: Jec =f( Je )...(13)

In one embodiment, the image edge detection method 10 can adjust the edge saliency degree Je to an appropriate value through a mapping function f(.), and the above-mentioned mapping method can be a simple linear mapping, a lookup table mapping, an exponential mapping, or a logarithmic mapping.

When the mapping method is simple linear mapping, the edge data Jec can be expressed as: Jec =min(max( a ( Je - b ),0),1)...(14)

When the mapping method is look-up table mapping, it can be performed according to a look-up table corresponding to the curve shown in Figure 3, and the edge saliency Je is clipped to a certain range before performing look-up table mapping. In this case, the edge data Jec can be expressed as: Jec =LUT(min(max(Je,a),b))...(15)

In practice, the processing circuit 210 can be used to perform normalization and mapping processing on the edge saliency degree Je. In one embodiment, the processing circuit 210 includes a table lookup circuit, and the table lookup circuit completes the mapping process by looking up a lookup table stored in memory in advance.

It is worth noting that the image edge detection method 10 according to the embodiment of the present invention can select different mapping methods to adjust the edge saliency degree Je according to different application scenarios. For example, when the sensitivity (ISO value) of the image I is a low sensitivity (such as ISO 100), the lookup table circuit of the processing circuit 210 performs mapping processing by looking up a first lookup table, and when the sensitivity (ISO value) of the image I is a high sensitivity (such as ISO 3200), the table lookup circuit of the processing circuit 210 performs mapping processing by looking up a second lookup table, and the first lookup table and the second lookup table correspond to different mapping curves.

Please refer to FIG. 4, which is a schematic diagram of an application scenario of an embodiment of the present invention. The edge data Jec of the image I pixels generated by the image edge detection method 10 according to the embodiment of the present invention can be applied to a sharpening process on the image I.

The sharpening process is used to improve the sharpness of the image. Generally speaking, the sharpening process is performed on the Y channel (ie, the brightness channel) of the YUV image. As shown in Figure 4, when image I is processed, the Y channel is first separated. Next, the details of the Y channel are extracted to generate a texture layer, an edge layer and a base layer, and the edge saliency Je generated by the image edge detection method 10 according to the embodiment of the present invention can be used as a parameter for adjusting the edge layer of the image. In practice, the edge layer of the image is adjusted by using the edge data Jec obtained after the edge saliency degree Je is normalized and mapped.

In one embodiment, the edge data Jec can be multiplied by the corresponding edge layer (as shown in FIG. 4 , that is, the edge data Jec is a multiplication coefficient of the edge layer). In other embodiments, the edge data Jec can also adjust the edge layer by logical operations such as addition, subtraction, or division. Finally, the image edge layer adjusted by the edge data Jec of the embodiment of the present invention is combined with the base layer to obtain a sharpened Y channel, which is then combined with the color-processed UV channel to output a processed YUV image.

Please refer to FIG. 5 , which is a schematic diagram of an image edge detection device 50 according to an embodiment of the present invention. The image edge detection device 50 can also be used to implement the image edge detection method 10 . The image edge detection device 50 includes a convolution circuit 502 , a product circuit 504 , a statistical circuit 506 and a judgment circuit 508 , wherein the convolution circuit 502 includes convolution circuit units 5022 , 5024 .

As shown in FIG. 5, the image I is input to the image edge detection device 50, and the image I can be convoluted through the convolution circuit units 5022 and 5024 of the convolution circuit (for example, the convolution circuit unit 5022 performs convolution operation on the image I with the x-direction gradient operator Kx, and the convolution circuit unit 5024 performs convolution operation on the image I with the y-direction gradient operator Ky), to obtain the first direction gradient data Gx and the second direction gradient data Gy. In order to meet the needs of the product circuit 504 and/or reduce the amount of calculation, the convolution circuit 502 may include a plurality of shift circuits to perform shift processing on the outputs of the convolution circuit units 5022, 5024 (for example For example, shift right by one bit).

Next, the product circuit 504 uses the square circuit and the multiplication circuit to calculate the square Gxx of the first direction gradient data Gx, the square Gyy of the second direction gradient data Gy, and the product Gxy of the first direction gradient data Gx and the second direction gradient data Gy, and the shift circuit and the truncation circuit Clip[. ] shift the result to the right by 2 bits, and intercept -2048 to 2047 bits to meet the needs of the statistical circuit 506 and/or reduce the amount of calculation.

Further, the statistical circuit 506 includes a summation circuit Σ (.) and a truncation circuit Clip[. ], to calculate the gradient statistics Ixx (i, j), Iyy (i, j), Ixy (i, j) in the local area according to the gradient data of the adjacent area adjacent to the target pixel A and perform truncation processing.

Finally, the judging circuit 508 can be a logic operation circuit, a shift circuit, a truncation circuit Clip[. ] and the division circuit realize the above formula (12) to obtain the corresponding edge salience degree Je. In one embodiment, the image edge detection device 50 may further include a processing circuit (not shown) for normalizing and mapping the edge saliency Je to meet the requirements of the application scene.

It should be noted that the number of bits indicated in FIG. 5 is used to illustrate the bit relationship between the various circuits of the image edge detection device 50 , rather than to limit the implementation of the image edge detection device 50 . In addition, the hardware circuit and its connections in the image edge detection device 50 are only a circuit implementation for implementing the image edge detection method 10 , and are not limited thereto.

In summary, the image edge detection method and the image edge detection device provided by the present invention judge the edge of the image with the gradient statistical information adjacent to the target pixel, thereby effectively improving the detection The accuracy of oriented edges in the image.

The above descriptions are only preferred embodiments of the present invention, and all equivalent changes and modifications made according to the scope of the patent application of the present invention shall fall within the scope of the present invention.

10: Image edge detection method

102-112: Steps

Claims (17)

An image edge detection method, executed on an image edge detection device, for processing an image comprising a plurality of pixels, the method includes: using a gradient operator in a first direction and a gradient operator in a second direction to perform a convolution operation on the image respectively to obtain gradient data in a first direction and gradient data in a second direction, wherein the first direction is perpendicular to the second direction; according to the gradient data in the first direction and the gradient data in the second direction, calculate gradient statistics in an adjacent area of a target pixel of the image to obtain a gradient statistic; A statistic that determines the saliency of an edge corresponding to the target pixel. The image edge detection method as described in Claim 1, wherein the operation of obtaining the gradient statistics includes: calculating the square of the gradient data in the first direction, the square of the gradient data in the second direction, and the product of the gradient data in the first direction and the gradient data in the second direction, so as to obtain the gradient statistics. The image edge detection method according to claim 1, wherein the operation of obtaining the gradient statistics includes: obtaining the gradient statistics according to a weighting coefficient, the gradient data in the first direction, and the gradient data in the second direction. The image edge detection method according to claim 1, wherein the adjacent area includes the target pixel and at least one adjacent pixel of the target pixel. The image edge detection method according to claim 1, wherein the edge saliency is determined according to a covariance matrix of the gradient statistics. The image edge detection method according to claim 5, wherein the greater the difference between the eigenvalue of the covariance matrix in the main direction and the eigenvalue in the primary direction, the greater the saliency of the edge of the target pixel. The image edge detection method according to claim 6, wherein the main direction and the secondary direction are orthogonal to each other. The image edge detection method according to Claim 1 further includes: normalizing the edge saliency corresponding to the image, and determining edge data of the image in a mapping manner. The image edge detection method as described in Claim 8 further includes: performing a sharpening process on the image according to the edge data. An image edge detection device for processing an image containing a plurality of pixels, which includes: a convolution circuit, including a first convolution circuit unit and a second convolution circuit unit, used to perform a convolution operation on the image with a gradient operator in a first direction and a gradient operator in a second direction, respectively, to obtain a gradient data in a first direction and a gradient data in a second direction, wherein the first direction is perpendicular to the second direction; Gradient statistics are calculated in a neighboring region of a target pixel to obtain a gradient statistic; and a judging circuit is used to determine an edge saliency corresponding to the target pixel according to the gradient statistic. The image edge detection device according to claim 10, further comprising: a product circuit for calculating the square of the gradient data in the first direction, the square of the gradient data in the second direction, and a product of the gradient data in the first direction and the gradient data in the second direction. The image edge detection device according to claim 10, wherein the statistics circuit is used to obtain the gradient statistics according to a weighting coefficient, the gradient data in the first direction and the gradient data in the second direction. The image edge detection device according to claim 10, wherein the adjacent area includes the target pixel and at least one adjacent pixel of the target pixel. The image edge detection device according to claim 10, wherein the judging circuit determines the edge saliency according to a covariance matrix of the gradient statistics. The image edge detection device according to claim 14, wherein the greater the difference between the eigenvalue in the main direction and the eigenvalue in the primary direction of the covariance matrix, the greater the saliency of the edge of the target pixel. The image edge detection device as described in claim 15, wherein the main direction and the secondary direction orthogonal to each other. The image edge detection device according to claim 10 further includes: a processing circuit for normalizing and mapping the edge saliency corresponding to the image to obtain edge data of the image.

Images