WO2019041590A1

WO2019041590A1 - Edge detection method using arbitrary angle

Info

Publication number: WO2019041590A1
Application number: PCT/CN2017/112917
Authority: WO
Inventors: 刘苏; 张劭龙; 耿兴光; 张以涛; 张俊; 张海英
Original assignee: 中国科学院微电子研究所
Priority date: 2017-08-31
Filing date: 2017-11-24
Publication date: 2019-03-07
Also published as: CN109427066A; CN109427066B

Abstract

An edge detection method using an arbitrary angle comprises: determining a boundary of an edge detection angle range; determining an arbitrary angle in an edge detection angle range; performing a convolution operation on each of a number of constructed straight pixel lines and a first derivative of a Gaussian function, obtaining an absolute value of a convolution operation result, and obtaining a local greatest value of the absolute value; assigning a gray value to the obtained local greatest value, and setting a gray scale of other non-local greatest values as 0; replacing original image pixels with image pixels having the gray values; and performing superposition of the gray scale on a number of images obtained by means of different edge detection angles, setting, according to requirements on an actually required edge image, a binarization threshold for the gray scale of an image resulting from multiple times of superimposition, and performing, according to the threshold, binarization on the image, so as to obtain a desired edge. The method provides an algorithm for edge detection performed by using an arbitrary angle and reduces complexity of an image edge detection algorithm.

Description

Edge detection method at any angle

Technical field

The present invention relates to an image processing method, and more particularly to an edge detection method at an arbitrary angle.

Background technique

Image brings humanity an image of the world of thinking and is an important way for human beings to understand the world. The mutations that exist in the image and the discontinuous and unstable structures are called edges. The edges often carry a wealth of image information. These edge points constitute the contour of the object, and these contours are often of interest to the researcher. They focus on the characteristics of the research target, and are extremely important for subsequent image segmentation, image matching, target recognition, and computer vision. How to convert images with unclear outlines into clear edge images has become the direction that people have been studying intensively for many years. In decades of research, people have been introducing mathematical methods to extract and interpret image edges. From the original gradient-based Prewitt operator, Sobel operator, etc. to LoG operator and Canny operator, wavelet transform to machine learning, the depth and difficulty of edge detection problem are reflected.

The multi-angle edge detection algorithm based on the gradient principle uses a N*N gradient template to convolve the two-dimensional image. Since the template is generally square and its size is up to 5 pixels * 5 pixels, the template can generate a gradient direction of up to 16, ie 0 °, 30 °, 45 °, 60 °, 90 °, 120 °, 135 ° , 150°, 180°, 210°, 225°, 240°, 270°, 300°, 315°, and 330° directions. The classical two-dimensional wavelet transform modulus maximum edge detection method can only perform non-maximum value suppression according to the angle classification after finding the gradient along the x direction and the y direction. Therefore, using the existing angle edge detection method to perform arbitrary angle edge detection on the image edge basically relies on rotating image and rotating coordinates. However, when the image is rotated and the coordinates are rotated, the image is interpolated, which causes the image gray information to change. Therefore, the edge of the image is recognized after rotating the image and rotating the coordinates, and the edge of the image cannot be guaranteed. It is also necessary to rotate the edge image back to the original position according to the angle of rotation, which again causes the edge image information to change. In addition, rotating images and rotating coordinates can cause image size changes and image boundary problems that can increase image processing difficulty.

Summary of the invention

The present invention provides a method for realizing a single pixel arbitrary angle edge detection without changing image information.

Specifically, the present invention provides an edge detection method at any angle, including the following steps:

Obtaining a gray value of all the two-dimensional pixel points of the image to be detected, where the size of the image to be detected is n×m, where m and n are positive integers;

The gray value of the selected part of the above two-dimensional pixel points is extracted by the following rule:

(a) starting from the top leftmost pixel of the image to be detected, continuously selecting k pixels; wherein k is selected from a positive integer;

(b) successively select k pixels for each row, except that the starting position of each row is the end position of consecutive k pixels in the previous row, that is, according to the compact connection mode; or the end of consecutive k pixels in the previous row Add one position, that is, select according to the loose connection mode;

(c) When selecting from the second line, first select according to the i-th compact connection mode, and then select according to the j-time loose connection mode, so that the loop is r times, that is, the bottom line of the bottom of the image to be detected is reached; wherein, i , j, r are all selected from positive integers, and the user can implement edge detection of any extraction angle by setting i, j, r, k;

Convolution operation is performed on the matrix of several gray values stored in a matrix form and the first derivative f _σ (x) of the Gaussian function, and then the absolute value of the convolution operation is taken, and the absolute value is taken locally. maximum;

In the matrix corresponding to all the two-dimensional pixel points of the image to be detected, the obtained local maximum value position is assigned to a non-zero gray value, and the gray value of the other pixel positions is set to zero.

Wherein the first derivative f _σ (x) of the Gaussian function is

Where σ is a constant, and the value ranges from 1 to 10.

Wherein, the non-zero gray value is 255/the number of edge detection angles.

Wherein, after the step of assigning the obtained local maximum value to a non-zero gray value and the gray value of the other pixel position to 0, the method further comprises: replacing the pixel represented by the obtained gray value matrix with the original The step of corresponding pixels on the image.

Among them, the user performs 4 to 8 independent i, j, r, and k settings to achieve 4 to 8 different times. Extract the edge detection of the angle.

Wherein, a plurality of gray value matrices obtained by different edge detection angles are superimposed in an image display form, and a gray level threshold is set according to an actual required edge image requirement for the gray level of the plurality of superimposed images, according to the binary value The threshold is binarized to obtain the desired edge.

Among them, the desired edge obtained is a single pixel wide edge.

Based on the above technical solutions, the pulse recognition method of the present invention has the following beneficial effects as compared with the prior art:

1. The present invention provides an algorithm that can achieve edge detection at any angle in the range of [0°, 360°];

2. The invention can realize the edge detection of the [0°, 360°] angle interval by applying only the [45°, 90°] edge detection angle, and reduce the complexity of the image edge detection algorithm;

3. The present invention discloses for the first time a formula for constructing an arbitrary angle edge detection operator;

4. The edge detection angle construction method of the present invention is more achievable than the existing angle-based classical operator;

5. The algorithm transforms the two-dimensional image edge recognition problem into one-dimensional curve signal processing problem, which reduces the complexity of the algorithm;

6. The edge generated by this algorithm is a single pixel wide edge.

DRAWINGS

Figure 1 is a schematic illustration of a compact connection of k adjacent pixels of an image;

2 is a schematic diagram of a loose connection of k adjacent pixels of an image;

3 is a schematic diagram of an image edge detecting an arbitrary angle composition form of k adjacent pixels;

Figure 4 is a schematic illustration of a compact connection of two adjacent pixels of an image;

Figure 5 is a schematic illustration of a loose connection of two adjacent pixels of an image;

6 is a schematic diagram showing an arbitrary angle composition form of two adjacent pixels of an image edge detection;

Figure 7 is a schematic diagram of a portion of the image beyond the boundary of the image to complement the 0 amplification;

Figure 8 is a relational expression for superimposing a plurality of different detection angles;

Figure 9 is a schematic diagram of superimposing a plurality of different detection angles;

Figure 10 is an original view and a comparison chart of angle optimization and multiple angle superposition;

11 to 14 are the relationship between the number of angles of the circle, the circle and the letters, the circle, the circle and the letter, and the relationship between the connected domain and the number of pixels P;

Figure 15 is a schematic view of the break connection of the arm edge;

Figure 16 is an image of the arm and wrist edges;

Figure 17 is a curve of an arm wrist transformed into an edge of a one-dimensional curve and a filtered or high-order polynomial fit;

Figure 18 is a graph showing the curvature of the arm wrist and the corresponding curvature;

Figure 19 is an image of the arm wrist edge with radial artery information;

Figure 20 is a segmented radial artery image;

Figure 21 is a ordinate ordinate averaging and straight line fitting curve of the radial artery;

Fig. 22 is a coordinate display diagram of the radial artery.

Detailed ways

The present invention will be further described in detail below with reference to the specific embodiments of the invention,

The invention discloses an edge detection method of an arbitrary angle, which is obtained by acquiring gray values of an image to be detected, and then scanning the image by using pixel lines of different angles to respectively extract grays of pixels corresponding to the plurality of pixel lines. The degree value is stored as a matrix, and the obtained matrix is convoluted with the first derivative f _σ (x) of the Gaussian function, and the absolute value of the convolution operation result is taken, and the local maximum is taken for the absolute value. a value; in the matrix corresponding to all the two-dimensional pixel points of the image to be detected, the obtained local maximum value position is given a non-zero gray value, and the gray value of the other pixel position is set to 0, thereby obtaining a local pole Intermittent points or connections for large points. Those skilled in the art can perform interpolation or fitting based on these points or lines to obtain continuous line segments, and can also superimpose the results obtained by multiple pixel lines of different angles, and then binarize to obtain desired edges, and can also be based on further Connect the domain operations to find a continuous edge line of a single value.

Specifically, the edge detection method of any angle of the present invention includes the following steps:

(a) starting from the top leftmost pixel of the image to be detected, continuously selecting k pixels; wherein k is a positive integer greater than or equal to 1;

(b) successively select k pixels for each row, just the starting position of each row It is the end position of consecutive k pixels in the previous row, that is, according to the compact connection mode; or the end position of consecutive k pixels in the previous row is added, that is, according to the loose connection mode;

(c) Selecting according to the i-th compact connection mode, and then selecting according to j times of loose connection mode, so that the cycle is r times, that is, the bottom line of the bottom of the image to be detected is reached; wherein i, j, r are both It is a positive integer; according to the above settings, the following formula can be obtained:

Number of cycles r × (compact times i + loose times j) +1 = number of lines m;

Number of cycles r × compact number i × (number of pixels per line k-1) + number of pixels per line k × number of cycles r × loose times j + number of pixels per line k = number of columns n;

The thus obtained line segment which is obtained from the pixel at the upper leftmost corner of the image to be detected and which is bent at the lowermost row of the image to be detected is referred to as a "pixel line". The gray value of the pixel points covered by the straight line of the corresponding pixel is extracted by setting different k values each time by different extraction angles (also referred to as edge detection angles).

Convolution operation is performed on a plurality of pixel lines stored in a matrix form and a first derivative f _σ (x) of a Gaussian function, and the absolute value of the convolution operation result is taken, and the absolute value is taken as a local maximum Value; the first derivative fσ(x) of the Gaussian function is

Where σ is a constant, the value ranges from 1 to 10; the convolution operation formula is expressed as h _{n, σ} (θ)=g _n (θ)*f _σ (x), where g _n (θ) represents the line extraction according to the pixel The gray value matrix, h _{n, σ} (θ) represents the result of the convolution operation.

In the matrix corresponding to all the two-dimensional pixel points of the image to be detected, the obtained local maximum value position is assigned to a non-zero gray value, and the gray value of the other pixel positions is set to zero. Preferably, the non-zero gradation value is, for example, 255/the number of edge detection angles.

Preferably, a plurality of gray value matrices obtained by different edge detection angles may be grayscale superimposed in an image display form, and a binarization threshold is set according to an actual required edge image requirement for the gray scale of the image after multiple superpositions, according to the The binarization threshold binarizes the image to obtain the desired edge. The specific calculation method is shown in Figs. 8 and 9, for example, but Figures 8 and 9 are only schematic and are not intended to limit the present invention.

Preferably, different edge detection angles can be selected, for example, from 4 to 8. As shown in Figures 10 to 14, it has been experimentally verified that the best effect is obtained when different edge detection angles are selected from 4 to 8.

The step of extracting the gray value of the selected part of the two-dimensional pixel by using the above rule is based on the following principle:

The present invention defines two extraction modes, referred to as compact connections and loose connections, respectively:

As shown in FIG. 1, the compact connection means that the first extraction position of the next row of pixels is at the same position as the last extraction position of the pixel of the previous row, and the extracted gray value is represented by the matrix Q _θ2L as follows:

As shown in FIG. 2, the loose connection means that the first extraction position of the pixel of the next row is located one bit to the right of the last extraction position of the pixel of the previous row, that is, the position of one plus, and the gray value extracted by the matrix is Q. _{θ2L is} expressed as follows:

As shown in FIG. 3, the above-mentioned compact connection and loose connection can be mixed according to a certain rule, for example, i times compact connection, j times loose connection, and then repeated r times. The above i, j, and r are all positive integers not larger than the number of rows m.

In the present invention, the different extraction angles (edge detection angles) direction, in the numerical setting, are represented by the number of pixels extracted in each row, the number of rows, the number of repetitions of the compact connection and the loose connection, etc., which can be set by These parameters are used to determine the specific extraction angle with.

For example, for a compact connection, each row extracts two pixels (k=2) and repeats until the bottom row of the image to be detected (i=1, j=0, r=m-1), then it extracts the angular direction, ie Edge detection angle

That is 45°.

For another example, for a loose connection and a staggered arrangement of compact connections (i=1, j=1), two pixels (k=2) are extracted for each line, and are repeated until the bottom line of the image to be detected (r=(m- 1) / 2), then extract the angular direction, that is, the edge detection angle

That is, 59°, which is approximately 60°.

The derivation process of the edge detection method of the present invention adapted to the detection of an arbitrary angle is as follows, wherein the number of consecutive extracted pixels per line is k=2.

(1) Construct an edge detection angle interval boundary

The adjacent pixel point relationship of the image is divided into a compact connection and a loose connection. Taking two pixels per line as an example, the compact connection is as shown in FIG. 4: starting from the pixel in the leftmost column of the image, and the pixel of the adjacent row. The first and last pixels are vertically connected, and each two lines form a compact connection unit. In this way, several compact connecting units are connected in a line up to the image boundary, and the angle between this line and its y-axis projection is the edge detection angle. Its matrix Q _{θ2L is} expressed as:

Edge detection angle

The loose connection is shown in Figure 5: starting from the top left corner of the image, starting with the pixels of the leftmost column and the top row, the pixels of the adjacent rows are connected diagonally to the first and last pixels, and each two rows form a loose connection unit. According to this method, a plurality of loose connecting units are connected in a line up to the image boundary, and the angle between the line and its y-axis projection is the edge detecting direction. Its matrix Q _{θ2R is} expressed as:

Edge detection angle

Therefore, the edge detection angle composed of a two-pixel compact connection unit

Is the left boundary of the angular interval of the segment. Edge detection angle composed of two-pixel loose connection unit

Is the right border of the angular interval of the segment. Therefore, the angle interval is (θ _2L , θ _2R ).

When the number of pixels is k pixels, the left boundary of the edge detection angle interval composed of compact connecting units of k pixels

Where k=2,3,...

The right edge of the edge detection angle interval consisting of a compact connecting unit of k pixels

Where k=2,3,...

Therefore, the union of the detected angle interval boundaries is (θ ₁ , θ ₂ ) ∪ (θ ₃ , θ ₄ ) ∪ ... ∪ (θ _n-1 , θ _n ); the range of the union is (45°, 90°) ).

(2) Construct any angle in the edge detection angle interval

Taking a unit connected by two pixels as an example, the arbitrary angles in the interval are composed as follows:

As shown in Fig. 6, i compact connections and j loose connections constitute one unit repeated r times, and the relationship between the number of rows m and the number of columns n and i, j and r is:

r(i+j)+1=m; (1)

Ri(k-1)+krj+k=n; (2)

Therefore the edge detection angle in each angular interval

In addition, each boundary condition also conforms to the above formula.

Therefore, by the above method, the pixels in the image can be combined according to the required angle, and the image is complement-zero amplified for the part of the algorithm that realizes the boundary beyond the image, as shown in FIG. 7 .

(3) Convolution operation of a plurality of pixel lines constructed by the above method with a first derivative f _σ (t) of a Gaussian function, and taking an absolute value of the convolution operation result, and taking a local maximum value for the absolute value; upper bound angle to the edge detection compact connecting a number of pixels to explain an example of generating a straight line, as a starting point to generate the left-side boundary straight line a number of pixels _{_{_{X 1, X 2 ... X m}}} , a number of pixels or more straight lateral boundary is generated starting Y ₁ ...Y _m-1 ; where m is the row and k is the number of connected pixels.

Taking the edge detection lower boundary angle generated by the loose connection as an example to explain a number of pixel lines, the left side boundary is used as the starting point to generate a number of pixel lines X' ₁ , X′ ₂ ... X′ _m , and the upper side boundary is the starting point. Y' ₁ ... Y'_m-1; where m is a row and k is the number of connected pixels.

Each pixel line is convoluted with the first derivative f _σ (t) of the Gaussian function, and the absolute value of the convolution operation is obtained: |f _σ (t)*X ₁ |, |f _σ (t )*X ₂ |,...|f _σ (t)*X _m ||f _σ (t)*X′ ₁ |,|f _σ (t)*X′ ₂ |,...|f _σ (t)*X ' _m | and |f _σ (t)*Y ₁ |,...|f _σ (t)*Y _m-1 ||f _σ (t)*Y′ ₁ |,...|f _σ (t)*Y′ _M-1 |. The edge detection angle is reduced from [0°, 360°] to [0°, 180°] by convolving and constructing absolute values of the constructed pixel lines. Therefore, it is only necessary to process the image in the interval of the edge detection angle [0°, 180°].

(5) grading the gradations of several images obtained by different edge detection angle directions, setting a binarization threshold for the gradation of the image after multiple superimposition according to the actual required edge image requirement, and performing two images according to the binarization threshold Value processing. The final edge is obtained.

The edge detection angle range is (45°, 90°), and the 45° edge detection angle is a pixel line in which one pixel is sequentially connected, that is, when k=1. The 90° direction is a vertically segmented image, and each column of pixels constitutes a pixel line. Therefore, the detection angle range [45°, 90°] can be achieved.

The angle range [45°, 90°] can be mapped to [0°, 45°], [90°, 135°] and [135°, 180°] by transposing and flipping the image matrix. The specific method is as follows:

The image matrix is flipped horizontally, and the edge detection angle interval is mapped from [45°, 90°] to [90°, 135°]. After the image matrix is transposed, the edge detection angle interval is mapped from [45°, 90°] to [135°, 180°]. After the image matrix is horizontally flipped and transposed, the edge detection angle interval is mapped from [45°, 90°] to [0°, 45°]. Based on the above method, the edge detection of the [0°, 360°] angle interval can be realized only by applying the [45°, 90°] edge detection angle.

application

The edge recognition method of any angle of the present invention can be applied to the pulse recognition, and the pulse recognition method includes, for example, the following steps:

1. Identify the edges of the arms and wrists to be tested to create the edge lines of the arms and wrists. The algorithm for recognizing the edges of the arms and wrists is the edge detection algorithm of any angle of the present application.

2. Pre-treatment of the edges of the arms and wrists to further optimize the edges of the arms and wrists to provide protection for subsequent wrist veins. This step specifically includes identifying the maximum connected domain of the arm edge, the arm edge breakpoint connection, and the arm wrist curve fitting as follows:

(1) Identify the maximum connected domain of the arm edge: identify the connected domain of the generated edge image, and find the largest connected domain of the right edge of the image. If the maximum connected domain runs through the left and right borders of the image, that is, there is no breakpoint in the connected domain, the maximum connected domain can be considered as the edge of the arm wrist.

(2) Arm edge breakpoint connection: Connect the arm edge segments to form an integral edge of the arm wrist that runs through the left and right borders of the image. In the case of a breakpoint at the edge, the maximum connected domain is only a part of the edge of the arm's wrist, so the other arm wrist edge segments need to be connected. The edge segment is found in the range of 2 pixels in the upper, upper left, left, lower left, and lower directions from the left breakpoint of the largest connected domain. If there are other connected domains in the search range, the two connected domains are connected, and the intermediate breakpoint pixels complement the pixels between the two segments by interpolation or other fitting, eventually forming a new connected domain and The breakpoint on the left side of the connected domain is the origin and further search for other edge segments until reaching the left edge of the image.

(3) Curve fitting of the arm wrist: Eliminate the step point generated in the process of turning the edge of the two-dimensional image into a one-dimensional curve, so that the one-dimensional arm edge curve of the transformation is smoother, and the edge feature of the arm wrist is highlighted.

3. Identifying the sacral stalk algorithm is used to identify the sacral stalk feature points: firstly extract the feature of the extracted arm wrist edge, identify the depression between the hand and the sacral stem, and find the lowest point of the depression. The curvature of the humeral stem at the top of the epidermis is characterized by the fact that the wrist is sunken to the arm with a maximum curvature point, that is, the point where the boundary changes to a greater extent. Second, look for peaks and valleys from the maximum curvature curve near the depression. Finally, it is recognized whether there is a peak at the arm edge near the peak of the curvature curve. If there is a peak, the point can be identified as the off-axis x coordinate; if there is no peak, the curve peak and valley at that point is identified as the off-axis x coordinate.

4. Radial artery image segmentation and vein recognition are used to segment the radial artery image and fit into a linear function that reflects the trend of the radial artery. The specific steps include:

(1) Area construction and threshold settings for providing a threshold reference for the binarized brachial artery.

(2) Binaryized radial artery region for separating the radial artery image from other images.

(3) The radial artery straight line fitting is used to obtain a linear function reflecting the trend of the radial artery and the final Guanmai coordinates.

In one embodiment, the pulse recognition method comprises the following steps:

First, the entire image is performed by using the above-described edge detection method of any angle of the present application. Edge recognition creates continuous or interrupted points and/or lines at the edge of the arm's wrist.

The arm edge is then pre-treated to further optimize the edge of the arm wrist to provide protection for subsequent wrist pulse recognition. The pre-processing process includes identifying the largest connected domain of the arm edge, the breakpoint connection of the arm edge, and the curve fitting of the arm wrist.

1) Identify the maximum connected domain of the arm edge: identify the connected domain of the generated edge image and find the largest connected domain on the right edge of the image. If the maximum connected domain runs through the left and right borders of the image, that is, there is no breakpoint in the connected domain, the maximum connected domain can be considered as the edge of the arm wrist.

2) As shown in Figure 15, the arm edge breakpoint connection includes the steps of joining the arm edge segments to form an integral edge of the arm wrist that runs through the left and right borders of the image. In the case of a breakpoint at the edge, the maximum connected domain is only a part of the edge of the arm's wrist, so the other arm wrist edge segments need to be connected. The edge segment is found in the range of 2 pixels in the upper, upper left, left, lower left, and lower directions from the left breakpoint of the largest connected domain. If there are other connected domains in the search range, the two connected domains are connected, and the intermediate breakpoint pixels complement the pixels between the two segments by interpolation or other fitting, eventually forming a new connected domain and The breakpoint on the left side of the connected domain is the origin and further search for other edge segments until reaching the left edge of the image.

3) As shown in Fig. 16 and Fig. 17, the wrist wrist curve fitting includes the following steps: using a low-pass filter or a polynomial curve fitting to eliminate the step point generated in the process of converting the edge of the two-dimensional image into a one-dimensional curve, so that The converted one-dimensional arm edge curve is smoother, highlighting the edge features of the arm wrist.

The sacral stem algorithm is used to identify the characteristic points of the sacral stem. As shown in Fig. 18, the extracted wrist arm edge is first extracted, and the depression between the hand and the sacral stem is identified to find the lowest point of the depression. The curvature of the humeral stem at the top of the epidermis is characterized by the fact that the wrist is sunken to the arm with a maximum curvature point, that is, the point where the boundary changes to a greater extent. Second, look for peaks and valleys from the maximum curvature curve near the depression. Finally, it is recognized whether there is a peak at the arm edge near the peak of the curvature curve. If there is a peak, the point can be identified as the off-axis x coordinate; if there is no peak, the curve peak and valley at that point is identified as the off-axis x coordinate.

Radial artery image segmentation and pulse recognition. An area is constructed with each pixel in the previously generated edge image (Fig. 19) as the origin. The threshold of the mean and the variance is set according to the statistical rule of the mean and variance of the region of the radial artery boundary position. Calculate the mean and variance of the pixels in each edge pixel area. The mean and variance of the pixels in each of the generated edge pixel regions are successively compared with the threshold, and the region meeting the threshold condition is binarized (Fig. 20). Instigation of binarization The pulse image is averaged on the ordinate of the pixel to obtain a curve describing the radial artery image. A quadratic polynomial straight line fitting is performed on the curve to obtain a linear function including the trend of the radial artery (Fig. 21), and the x-coordinate of the pulse is substituted into the linear function to obtain the ordinate of the pulse. The position of the pulse in the image can be determined, as shown in Figure 22.

The specific embodiments of the present invention have been described in detail with reference to the preferred embodiments of the present disclosure. All modifications, equivalents, improvements, etc., made within the spirit and scope of the present disclosure are intended to be included within the scope of the present disclosure.

Claims

An edge detection method at any angle includes the following steps:

Obtaining a gray value of all the two-dimensional pixel points of the image to be detected, where the size of the image to be detected is n×m, where m and n are positive integers;

The gray value of the selected part of the above two-dimensional pixel points is extracted by the following rule:

(a) starting from the top leftmost pixel of the image to be detected, continuously selecting k pixels; wherein k is selected from a positive integer;

(b) successively select k pixels for each row, except that the starting position of each row is the end position of consecutive k pixels in the previous row, that is, according to the compact connection mode; or the end of consecutive k pixels in the previous row Add one position, that is, select according to the loose connection mode;

(c) When selecting from the second line, first select according to the i-th compact connection mode, and then select according to the j-time loose connection mode, so that the loop is r times, that is, the bottom line of the bottom of the image to be detected is reached; wherein, i , j, r are all selected from positive integers, and the user can implement edge detection of any extraction angle by setting i, j, r, k;

Convolution operation is performed on the matrix of several gray values stored in a matrix form and the first derivative f σ (x) of the Gaussian function, and then the absolute value of the convolution operation is taken, and the absolute value is taken locally. maximum;

In the matrix corresponding to all the two-dimensional pixel points of the image to be detected, the obtained local maximum value position is assigned to a non-zero gray value, and the gray value of the other pixel positions is set to zero.
The method of claim 1 wherein the first derivative f σ (x) of the Gaussian function is
Where σ is a constant, and the value ranges from 1 to 10.
The method of claim 1 wherein said non-zero gray value is 255/the number of edge detection angles.
The method according to claim 1, wherein after the step of assigning the obtained local maximum value to a non-zero gray value and the gray value of the other pixel position is set to 0, The pixel represented by the gray value matrix replaces the corresponding pixel on the original image A step of.
The method according to claim 1, characterized in that the user performs 4 to 8 independent settings of i, j, r, k to realize edge detection of 4 to 8 different extraction angles.
The method according to claim 5, wherein the plurality of gray value matrices obtained by different edge detection angles are gray-scale superimposed in an image display form, and the gray scales of the plurality of superimposed images are required according to actual required edge image requirements. A binarization threshold is set, and the image is binarized according to the binarization threshold to obtain a desired edge.
The method of claim 6 wherein the desired edge obtained is a single pixel wide edge.