CN107218894B - Rapid and stable sub-pixel precision device thickness detection method - Google Patents
Rapid and stable sub-pixel precision device thickness detection method Download PDFInfo
- Publication number
- CN107218894B CN107218894B CN201710295849.XA CN201710295849A CN107218894B CN 107218894 B CN107218894 B CN 107218894B CN 201710295849 A CN201710295849 A CN 201710295849A CN 107218894 B CN107218894 B CN 107218894B
- Authority
- CN
- China
- Prior art keywords
- boundary
- brightness
- curve
- pixels
- positioning
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01B—MEASURING LENGTH, THICKNESS OR SIMILAR LINEAR DIMENSIONS; MEASURING ANGLES; MEASURING AREAS; MEASURING IRREGULARITIES OF SURFACES OR CONTOURS
- G01B11/00—Measuring arrangements characterised by the use of optical techniques
- G01B11/02—Measuring arrangements characterised by the use of optical techniques for measuring length, width or thickness
- G01B11/06—Measuring arrangements characterised by the use of optical techniques for measuring length, width or thickness for measuring thickness ; e.g. of sheet material
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
- Length Measuring Devices By Optical Means (AREA)
Abstract
The invention relates to a method for quickly and stably detecting the thickness of a sub-pixel precision device. The method comprises the following steps: preliminary boundary positioning: dividing the image into a target area and an air area, and then obtaining a boundary of the two areas as a primary boundary positioning result; and (3) sub-pixel accurate positioning: solving by adopting an arc tangent curve fitting method and a particle swarm optimization algorithm, and fitting by adopting a Bezier curve according to the positioning result of each column; and (5) processing an abnormal condition. The method is improved aiming at the sub-pixel boundary positioning algorithm, and further realizes the speed improvement of the algorithm on the basis of ensuring that the positioning precision meets the requirement from the three aspects of initial positioning, approximate parameter estimation and rapid optimal parameter search, thereby improving the application potential of the sub-pixel boundary positioning algorithm.
Description
Technical Field
The invention belongs to the field of automatic detection, relates to the technical field of industrial vision, and particularly relates to a method for quickly and stably detecting the thickness of a sub-pixel precision device.
Background
With the rapid development of the manufacturing industry in China, the importance on the quality is increasingly important. In the production process of various products, various components and parts are subjected to quality detection by using automatic equipment, so that the rejection rate can be reduced, the loss of materials is reduced, and the method becomes a powerful means for improving enterprise profits and brand influence. Industrial vision techniques are the most commonly used techniques in the field of automated inspection.
The measurement of the thickness and size of components is the most common application of industrial vision inspection. After the size parameters of the components are obtained, whether the components are qualified or not can be judged, so that the components are removed in advance, and the influence on the subsequent manufacturing links is prevented. By measuring various size parameters of the components and comparing the parameters with standard sizes, the components which are unqualified, and the unqualified proportion and details can be known.
Thickness detection of devices is the most common step in automated detection. Thickness detection is currently achieved by locating the sub-pixel boundaries of the device in the image. However, the current common sub-pixel boundary positioning algorithm has high complexity and high-degree redundant calculation, so that the efficiency of automatic detection equipment is reduced, the capacity is reduced, and the profit of an enterprise is influenced.
Disclosure of Invention
The invention aims to provide a fast and stable method for detecting the thickness of a sub-pixel-level precision device, which is used for improving a sub-pixel boundary positioning algorithm, and further realizing the speed improvement of the algorithm on the basis of ensuring that the positioning precision meets the requirement from three aspects of initial positioning, approximate parameter estimation and fast optimal parameter search, thereby improving the application potential of the sub-pixel boundary positioning algorithm.
In order to achieve the above purpose, the invention adopts the technical scheme that: a method for detecting the thickness of a fast and stable sub-pixel precision device comprises the following steps:
1) preliminary boundary positioning: dividing the image into a target area and an air area, and then obtaining a boundary of the two areas as a primary boundary positioning result;
2) and (3) sub-pixel accurate positioning: solving by adopting an arc tangent curve fitting method and a particle swarm optimization algorithm, and fitting by adopting a Bezier curve according to the positioning result of each column;
3) and (5) processing an abnormal condition.
Further, the deviation of the initial positioning position of the boundary initial positioning and the final sub-pixel scenic spot coordinate is within the distance of 1 pixel.
Still further, the preliminary boundary positioning includes the following steps:
a) coarse to fine boundary preliminary positioning
Obtaining a preliminary positioning coordinate point of the interface according to the following process:
1a) image reduction: firstly, reducing the image to one fourth of the original size;
2a) extracting a target area reference brightness value and an air area reference brightness value: pre-calculating brightness reference values of a target area and an air area in an image, dividing all pixels into two large classes according to brightness values by adopting a clustering method, and then calculating the brightness average value of each class to be respectively used as the reference values of the target brightness and the air area brightness;
3a) and (3) binarization operation: after reference brightness values of a target area and an air area are obtained, the mean value of the reference brightness values and the air area is used as a binarization threshold value of the whole image, and binarization operation is carried out on all pixels;
4a) extracting a maximum connected region: extracting a maximum connected region in a binarization operation result;
5a) preliminary boundary localization based on symmetry; obtaining a preliminary judgment by utilizing the symmetry of a pixel brightness change rule curve near the boundary;
6a) exception handling during initial positioning;
b) location of load bearing interface
1b) Setting the range of the inclination angle, traversing the angle, and horizontally projecting the binary image according to the set angle value to obtain a projection curve;
2b) analyzing projection curves under various angles, counting the number of lines required to be experienced when the projection value is increased from 0 to an image width value N, and selecting an angle value corresponding to the condition of the least number of lines as an inclination angle of a bearing interface;
3b) finding a projection curve corresponding to the inclination angle, carrying out difference calculation on the projection curve, calculating the difference between the upper and lower projection values, and selecting the line with the maximum difference value as the position of the bearing interface;
4b) then, filtering out a value corresponding to the bearing interface in the differential curve, and searching the maximum differential value to be used as an initial positioning position of the top boundary of the device;
c) positioning of left and right boundaries of the device:
1c) obtaining a boundary line between the top boundary position of the device and the bearing interface position by using the top boundary position of the device obtained in the previous step and the bearing interface position;
2c) extracting points of which the coordinate values are positioned above the boundary line from all the boundary points;
3c) finding the leftmost side and the rightmost side according to the extracted points;
4c) vertically projecting the binary image within 20 pixels around the left point;
5c) differentiating the projection curve, and finding out the position with the maximum differential value as the left boundary of the device;
6c) within the range of 20 pixels around the right side point, vertically projecting the binary image, differentiating the projection curve, and finding out the position with the maximum differential value to obtain the right boundary of the device;
d) positioning of device top coordinate fillet
1d) Extracting the sub-outline corresponding to the device by using the bearing surface coordinate, the top coordinate of the device and the left and right boundary positions of the device obtained in the previous step;
2d) calculating a convex hull corresponding to the device sub-outline;
3d) generating a new binary image by using the convex hull, wherein the convex hull area is filled to be white;
4d) the convex hull area image is differenced with the original device image;
5d) in the difference map, four connected regions with the largest area are positioned;
6d) and calculating the positions of the four communication areas to obtain the fillet positions of the device.
Further, the sub-pixel accurate positioning is performed by fitting an arctangent curve to a column pixel curve near the boundary position, and the mathematical expression of the arctangent curve is as follows: y p1 atan (p2 x + p3) + p4, an accurate mathematical model of the luminance curve of the current column is obtained, then using the following formula: x0 ═ p3/p2, and the sub-pixel coordinates of each column were calculated.
Still further, the sub-pixel accurate positioning comprises the following steps:
a) direct estimation of two parameters, p1 and p 4: two parameters p1 and p4 are calculated in advance by using the shape of the luminance curve, and the two calculated parameter values are used as initial values of p1 and p 4:
the method comprises the following steps:
1a) gradually adding the pixels of the current column into the set from the upper part of the initial boundary position, if the standard deviation of the brightness of the pixels in the set is less than 5, continuing to add new pixels, if the standard deviation of the brightness of the pixels in the set is more than 5, stopping adding new pixels, and calculating the average value of the brightness of the pixels in the set at the moment as the average brightness of the upper side;
2a) gradually adding pixels of a current column into the set from the lower part of the initial boundary position, if the standard deviation of the brightness of the pixels in the set is less than 5, continuing to add new pixels, if the standard deviation of the brightness of the pixels in the set is more than 5, stopping adding new pixels, calculating the average value of the brightness of the pixels in the set at the moment to obtain the average brightness of the lower side, wherein the difference value between the average brightness of the upper side and the average brightness of the lower side is p 1;
3a) calculation of p 4: calculating the average value of all pixel brightness values on the brightness curve to obtain the vertical coordinate of the center point position of the boundary curve;
b) solving an arctangent curve based on a particle swarm algorithm:
the method comprises the following steps:
1b) setting a group of particles, 20 in total, each particle containing 4 parameters, and randomly setting an initial value to represent a solution of an arctangent curve;
2b) calculating the error between the curve corresponding to the current 20 particles and the actual brightness curve, and negating the error to serve as fitness;
3b) finding the most suitable curve parameter at present, then adjusting the direction and position of each particle, and continuing to calculate;
4b) repeating the steps 2b) and 3b) until the error of the best particle is less than the threshold value 10;
5b) calculating the accurate sub-pixel boundary position according to the formula x0 ═ p3/p 2;
c) application of the results of the previous column, initial values for p2 and p3 were obtained:
after the curve fitting for a column is finished, taking the values of p2 and p3 in the parameters of the column as the parameters of the brightness curve of the next column, and estimating p1 and p4 by calculation;
d) and (3) solving noise interference:
determining whether the fitting result of the current column is correct or not by judging the error in the fitting process, wherein for the situation of larger error, the column does not carry out accurate positioning of the boundary, and the column is repaired by smoothing the boundary at the later stage;
e) and (3) boundary smoothing:
and fitting the coordinates of the boundary points by adopting a Bezier curve fitting method.
Further, the abnormal condition processing comprises
a) No device warning: detecting by analyzing a vertical coordinate on a glass interface, fitting the coordinate of the interface into a straight line, and then calculating the distance from each point to the straight line, wherein if the distance from all the points to the straight line is less than 3 pixels, no device exists, and early warning is needed when no device exists in an image;
b) early warning of a plurality of devices: the method comprises the following steps:
1b) the plurality of devices are separated from each other: if the rising times and the falling times exceed 1 time, early warning is carried out;
2b) multiple pieces of devices overlap: after left and right boundaries of the device are obtained by utilizing the ascending and descending rules of the device coordinate, the thickness distribution of the device between the left and right boundaries is considered, and if the thickness value of a certain position between the left and right boundaries is equal to or more than the height of twice the thickness, the overlapping of a plurality of devices is indicated to be generated, and early warning is carried out;
c) gap early warning:
1c) contour extraction of the device region: in the preliminary positioning process of the device boundary, the positions of the left and right boundaries of the device are obtained, a binarization subimage corresponding to the device is intercepted according to the positions, then the outline is extracted, and the subsequent steps are to detect the gap on the basis of the outline;
2c) and (3) filtering the contours of the top and the left and right sides of the device: obtaining coordinates of the lower left corner and the lower right corner of the device by using the coordinates of the bearing interface obtained in the initial positioning process and the coordinates of the left and right boundaries of the device, and then filtering the top contour, the left side contour and the right side contour of the device according to the two coordinates to only leave contour points of a gap area;
3c) gap area size measurement: after the processing of the two steps 1c) and 2c), the remained straight line segment is the profile corresponding to the gap between the device and the glass plane, the profile point at the moment is analyzed, the boundary of the region is calculated, if the height of the region is less than 5 pixels and the width is more than 20 pixels, the region is the gap, and early warning is needed;
d) filtration of dust interference:
1d) and performing local straight line fitting on the boundary coordinates of the sub-pixels: the boundary points are divided into a plurality of handwriting combinations tentatively by adopting a mode of increasing and searching simultaneously;
2d) selecting coordinate points with the distance larger than a threshold value; in each subset, selecting out coordinate points with the linear distance from the contour point to be greater than a threshold value;
3d) segmenting the selected coordinate points; analyzing the coordinate points selected from the subsets, and dividing the coordinate points into a plurality of subsections according to whether the coordinate points are connected or not;
4d) the width of each subsection is inspected, and the width smaller than a threshold value is counted as dust;
5d) recalculating the coordinate points: filtering the sub-section corresponding to the dust, and replacing all coordinate points on the sub-section with coordinate points on a corresponding fitted straight line;
e) early warning of tailing phenomena
1e) And (3) judging the consistency of the vertical coordinates of the top boundary of the device: respectively selecting a region with the length of 20 at the left end and the right end of the device, then selecting a region with the length of 50 at the central position of the device, respectively calculating the average value of the vertical coordinates of the top contour points of the devices in the three regions, if one region exists in the left end and the right end, the average coordinate of the region is compared with the average coordinate of the central region, and the difference value is more than 5, then tailing phenomenon exists, and early warning is performed;
2e) and (3) analyzing the brightness consistency of the central region of the device and the regions at the head end and the tail end: as with 1e), selecting a region with the length of 20 at the left end and the right end of the device respectively, then selecting a region with the length of 50 at the center position of the device, calculating the average brightness of the three regions respectively, and if one region exists in the left end and the right end, the brightness of the region is different from the average brightness of the center region by more than 20, then tailing phenomenon exists, and early warning is carried out.
The invention has the technical effects that:
1. the method for positioning the boundary of the sub-pixel has the advantages that the method is high in speed, optimization is conducted to the maximum extent in each link of an algorithm, calculated amount is reduced to the minimum extent, and the overall operation speed of a system can be improved to the maximum extent;
2. the precision is high, and the deviation is within 3 micrometers from the actual thickness of the device (measured by using a vernier caliper for multiple times);
3. the method can adapt to various interference conditions, various abnormal conditions which are possibly met in the real use process of the system are processed, the normal detection is ensured not to be interfered by the abnormal conditions, and the early warning can be carried out on the abnormal conditions under the necessary condition.
Drawings
FIG. 1 is a general flow diagram of a device thickness detection system of the present invention;
FIG. 2 is a flow chart of the preliminary boundary positioning of the device of the present invention;
FIG. 3 is a flow chart of the positioning of the load bearing interface of the device of the present invention;
FIG. 4 is a flow chart of the positioning of the left and right boundaries of the device of the present invention;
FIG. 5 is a flowchart of the device fillet position calculation of the present invention;
FIG. 6 is a flow chart of sub-pixel boundary positioning according to the present invention;
FIG. 7 is a flowchart illustrating the estimation of the parameters p1 and p4 according to the present invention;
FIG. 8 is a flow chart of the PSO algorithm based arctangent curve solving of the present invention;
FIG. 9 is a flow chart of the abnormal situation handling process in the device thickness inspection process of the present invention;
FIG. 10 is a flow chart of the detection of the presence of multiple devices of the present invention;
FIG. 11 is a flow chart of the gap pre-warning of the device of the present invention;
FIG. 12 is a flow chart of dust filtration according to the present invention;
fig. 13 is a flow chart of the tailing phenomenon warning according to the present invention.
Detailed Description
Referring to the drawings, the invention deals with exceptional situations including: there is not the device in the field of vision, has a plurality of devices in the field of vision, and the overlap takes place for the device, has the gap between device and the glass quotation, and the dust on the device disturbs, and the device angle is put the tail flick phenomenon that inaccurately causes. The method is improved aiming at the sub-pixel boundary positioning algorithm, and further realizes the speed improvement of the algorithm on the basis of ensuring that the positioning precision meets the requirement from the three aspects of initial positioning, approximate parameter estimation and rapid optimal parameter search, thereby improving the application potential of the sub-pixel boundary positioning algorithm.
1. Preliminary positioning of boundaries
The image is divided into two regions, object and air, and then the boundary between the two regions is obtained as a preliminary boundary positioning result. Therefore, the subsequent sub-pixel level accurate positioning is carried out on the basis, and a large amount of searching time is saved. In addition, in the initial positioning stage, the horizontal position of the device is obtained according to the coordinates of the boundary points.
2. Sub-pixel accurate positioning
The invention adopts the method of arc tangent curve fitting. To achieve sub-pixel level accuracy location of the boundary points. In order to improve the speed of curve fitting, a particle swarm optimization algorithm is adopted for solving. Finally, aiming at the positioning result of each column, a Bezier curve is adopted for fitting, so that the final edge becomes smooth.
3. Abnormal situation handling
This step, needs to resolve some abnormal situations: there is not the device in the field of vision, has a plurality of devices in the field of vision, and the overlap takes place for the device, has the gap between device and the glass quotation, and the dust on the device disturbs, and the device angle is put the tail flick phenomenon that inaccurately causes.
Specifically, 1. position of the fast positioning device in the image: since the exact solution of the sub-pixel coordinates is performed for each coordinate point position, the calculation amount is large, and therefore, the initial positioning position of the boundary must be obtained before the sub-pixel coordinate solution. The deviation of the initial position from the final sub-pixel scene coordinates is within a distance of 1 pixel. Therefore, the search times can be reduced and the solution can be completed as soon as possible when the subsequent arctan curve is fitted.
a) Coarse to fine boundary preliminary positioning
Obtaining a preliminary positioning coordinate point of the interface according to the following process:
1a) image reduction
In order to improve the accuracy of coarse positioning, the image is firstly reduced to one fourth of the original size. Thus, the algorithm can process as few pixels as possible without affecting the final precision.
2a) Extracting a reference brightness value of a target area and a reference brightness value of an air area
Before the binarization operation is carried out, the brightness reference values of the target area and the air area in the image need to be calculated in advance. A clustering method is adopted, and for all pixels, the pixels are divided into two main classes according to the brightness values. Then, the average value of the brightness of each class is calculated and is respectively used as a reference value of the target brightness and the brightness of the air area.
3a) Binarization operation
And after the reference brightness values of the target area and the air area are obtained, the average value of the target area and the air area is used as a binarization threshold value of the whole image, and binarization operation is carried out on all pixels.
4a) Extracting maximum connected region
In order to reduce the noise interference as much as possible, the largest connected region in the binarization operation result needs to be extracted. Because the device and the bearing surface are adhered, the device and the bearing surface jointly form a target area, and therefore, the interference of various noises can be filtered only by finding a communication area with the largest area.
5a) Preliminary boundary localization based on symmetry
After the binarization operation, the boundary of the connected domain may have a certain deviation from the real boundary position, because the binarization threshold is not necessarily equal to the brightness value of the real boundary position. In order to obtain a sufficiently accurate initial boundary position and reduce the complexity of subsequent search, the invention obtains a preliminary judgment estimation by using the symmetry of a pixel brightness change rule curve near the boundary.
The method comprises the following steps: setting a column-direction search window with the length of 21 by taking the binary boundary as a central point; then, the 21 pixels are inspected one by one, and a row of pixels with the length of 11 are collected on each pixel; then, the central symmetry of the 11 pixels is analyzed; and selecting the position with the best symmetry as an initial positioning boundary.
6a) Exception handling during initial positioning
In the actually shot image, various abnormal conditions such as a gap between the device and the glass disk surface due to the round angle of the device or noise around the boundary due to binarization can be interfered. These interference situations can make the preliminary location of the boundary challenging.
The following method is adopted to realize boundary positioning under the interference condition: if the brightness of the pixels of the column changes normally (only one change exists), the situation is normal, and no interference exists; if the brightness of the pixel of the column jumps many times, the interference condition is present.
b) Location of load bearing interface
The load bearing interface is a platform for the device under test, generally capable of rotation, whose surface can be considered to be perfectly smooth. In order to obtain the accurate position of the device, the coordinates of the bearing interface need to be obtained in advance from the initial positioning boundary coordinates. In order to improve the stability and reduce the noise interference, the invention adopts a projection method to realize the positioning of the bearing interface. Meanwhile, in order to prevent the weighing wire from being inclined, the image needs to be rotated within a certain range and then projected.
The method comprises the following steps:
1b) setting the possible range of the inclination angle, and traversing the angle. According to the set angle value, carrying out horizontal projection on the binary image to obtain a projection curve;
2b) analyzing projection curves under various angles, counting the number of lines required to be experienced when the projection value is increased from 0 to an image width value N, and selecting an angle value corresponding to the condition of the least number of lines as an inclination angle of a bearing interface;
3b) finding a projection curve corresponding to the inclination angle, carrying out difference calculation on the projection curve, calculating the difference between the upper and lower projection values, and selecting the line with the maximum difference value as the position of the bearing interface;
4b) and then filtering out the value corresponding to the bearing interface in the differential curve, and searching the maximum differential value to be used as the initial positioning position of the top boundary of the device.
c) Positioning of left and right boundaries of a device
The left and right boundaries of the device also need to be determined in order to facilitate accurate positioning at the subsequent sub-pixel level. To reduce the noise interference, the boundaries are determined using a vertical projection. The method comprises the following specific steps:
1c) obtaining a boundary line (an average value of two vertical coordinates) of the top boundary position of the device and the position of the load-bearing interface obtained in the previous step;
2c) extracting points of which the coordinate values are positioned above the boundary line from all the boundary points;
3c) finding the leftmost side and the rightmost side according to the extracted points;
4c) vertically projecting the binary image within 20 pixels around the left point;
5c) differentiating the projection curve, and finding out the position with the maximum differential value as the left boundary of the device;
6c) in the same way, obtaining the right boundary of the device;
d) positioning of device top coordinate fillet
Because some devices have rounded corners, the boundary points have lower coordinate values than normal top boundary points. If the fillet position can not be accurately positioned, the subsequent accurate thickness measurement is influenced. The invention adopts the following method to position the fillet range:
1d) extracting the sub-outline corresponding to the device by using the bearing surface coordinate, the top coordinate of the device and the left and right boundary positions of the device obtained in the previous steps;
2d) calculating a convex hull corresponding to the device sub-outline;
3d) generating a new binary image by using the convex hull, wherein the convex hull area is filled to be white;
4d) the convex hull area image is differenced with the original device image;
5d) in the difference map, four connected regions with the largest area are positioned;
6d) and calculating the positions of the four communication areas, and calculating to obtain the fillet positions of the device.
2. Sub-pixel boundary localization
In an image captured by a high-definition industrial camera, the boundary of the device forms a staircase shape that gradually decreases in luminance. This shape is very similar to the shape of an arctan curve. The reason for this shape is that when an industrial camera captures an image, the boundaries of the device air cannot be separated clearly due to problems such as the accuracy of data sampling and the sharpness of the lens, and a certain degree of blurring occurs. Generally, the boundary with a certain transition effect needs to adopt a sub-pixel boundary positioning technology, otherwise, the measurement of the device size is influenced.
The traditional sub-pixel boundary positioning technology, such as a gray moment method, has large calculation amount, and influences the real-time performance of the algorithm. The invention realizes the sub-pixel positioning of the boundary by fitting the arc tangent curve and the column pixel curve near the boundary position. The mathematical expression for the arctangent curve is as follows:
y＝p1*atan(p2*x+p3)+p4 (1)
it is assumed that the column-direction image luminance curve near the boundary position also conforms to the shape of the arctan curve. Therefore, only equation (1) is needed to solve the four parameters of the luminance curve of the current column: p1, p2, p3, p4, an accurate mathematical model of the luminance curve of the current column can be obtained. Then, the precise sub-pixel boundary position, i.e. the position of the central point of the arctangent curve, can be obtained by using the following formula:
x0＝-p3/p2 (2)
from equation (2), the sub-pixel coordinates of each column can be calculated very quickly.
In order to improve the solving speed, the invention does not use an iterative solving method such as LM, but directly adopts Particle Swarm Optimization (PSO) to carry out fast search, thereby realizing the large-scale speed-up of the solving process.
The sub-pixel positioning process flow is as follows:
a) direct estimation of two parameters p1 and p4
The arctan curve has a total of four parameters, and in order to reduce the number of searches, two parameters, p1 and p4, may be calculated in advance using the shape of the luminance curve, so that the two calculated parameter values may be subsequently used as initial values of p1 and p 4.
Calculation of p 1-the physical quantity described by p1 is in fact the coverage of the arctan curve in the vertical direction. The estimate of this value can be obtained by calculating the average of the luminance on both sides of the boundary and then making the difference. However, when calculating the average value of the luminance on both sides, the luminance transition region needs to be filtered out so as not to cause interference. The method comprises the following steps:
1a) starting from above the initial boundary position, the pixels of the current column are added to the set gradually and successively. If the standard deviation of the luminance of the pixels in the set is less than 5, new pixels may continue to be added. If it is greater than 5, the addition of the new pixel is stopped. The average value of the luminances of the pixels in the set at this time is calculated as the average luminance on the upper side.
2a) Similarly, the same operation is performed from below the initial boundary position, and the lower average luminance is obtained. The difference between the two luminance values is p 1.
3a) Calculation of p 4-p 4 is actually the ordinate of the position of the center point of the boundary curve. The value is simple to calculate, and only the average value of all pixel brightness values on the brightness curve is needed.
b) Particle swarm algorithm (PSO) -based arc tangent curve solving
The arctangent curve model in the formula (1) contains 4 parameters, and cannot be accurately solved by using an equation solving method, so that the conventional method is to gradually approximate a better result by using an iterative method such as a gradient descent method. These methods tend to search more times and tend to fall into local optima. In order to solve the problems of speed and local optimization at the same time, the method adopts an optimization search strategy (PSO algorithm) to solve the arc tangent curve model.
Particle Swarm Optimization (PSO) is a classical evolutionary computation method. The algorithm simulates the law of birds looking for food. A flock of birds searches for food in an area, with their respective positions and velocities. If a bird finds food, other birds around the bird adjust their flight direction by referring to their direction. At the same time, each bird also has some memory, which remembers where food was the most so far, the direction of its flight, also to some extent with reference to that location. Thus, after a few iterations, the overall fitness of the bird population may reach a higher value, resulting in a better solution.
The specific method for solving the parameters of the arc tangent curve by using the PSO algorithm is as follows:
1b) setting a group of particles, 20 in total, each particle containing 4 parameters, and randomly setting an initial value to represent a solution of an arctangent curve;
2b) calculating the error between the curve corresponding to the current 20 particles and the actual brightness curve, and negating the error to serve as fitness;
3b) finding the most suitable curve parameter at present, then adjusting the direction and position of each particle, and continuing to calculate;
4b) repeating the two steps until the error of the best particle is less than the threshold value 10;
5b) the best 4 parameters of the particle at this time, which are the parameters of the arctangent curve that is finally needed, are calculated according to equation (2) to obtain the precise sub-pixel boundary position.
c) Application of the results of the previous column, initial values for p2 and p3 were obtained
After the curve fitting for specifying a certain column is completed, the values of p2 and p3 among the parameters of the column can be taken as the parameters of the luminance curve of the next column. While p1 and p4 are still estimated by calculation. This allows the initial value of the start to be closer to the true position when the current column is conducting a search of results, which further reduces the amount of searching.
d) Solution of noise interference problem
The presence of noise can cause the brightness curve to have a shape that differs greatly from the shape of a conventional arctan curve. At this time, the arctan curve model is directly applied to solve, and the final result is greatly different from the situation without noise interference. The invention determines whether the fitting result of the current column is correct or not by judging the error in the fitting process. For the case of large errors, the column does not perform precise location of the boundary, but is repaired later by boundary smoothing.
e) Boundary smoothing
And smoothing the preliminarily obtained boundary sub-pixel coordinate values and the coordinates between adjacent columns to reduce noise interference or the boundary jitter phenomenon caused by random deviation in the image sampling process. If the boundary coordinates are not smoothed, the subsequent dust filtering process may be affected, so that many real dusts cannot be detected, and the measurement result is affected.
The invention adopts a Bezier curve fitting method to complete the fitting of the boundary point coordinates.
3. Abnormal situation handling
a) Device-less warning
The strategy for detecting whether the device exists in the visual field is simple, but attention needs to be paid to saving time and improving algorithm efficiency. The invention adopts the following method to quickly verify whether the device exists in the visual field:
in the absence of a device, the glass interfaces in the image are straight lines. The detection can be carried out by analyzing the vertical coordinate on the interface of the glass. And fitting the coordinates of the interface into a straight line. The distance of each point to the line is then calculated. If all the points to straight line distances are small enough, there are no devices.
For the situation that no device exists in the image, early warning is needed to prevent the measurement result from being abnormal.
b) Multiple device early warning
Multiple devices may be present in the image simultaneously due to possible anomalies in the mechanical parts of the system. This situation must be pre-warned to prevent anomalies in the measurement results. The treatment is divided into two cases:
1b) multiple pieces of devices separated from each other
And (4) considering the rising and falling times of the boundary coordinates, and if the rising and falling times exceed 1 time, understanding early warning.
2b) Multiple pieces of devices overlap
After the left and right boundaries of the device are obtained by using the ascending and descending rules of the device coordinates, the thickness distribution of the device between the left and right boundaries needs to be considered. Since the thickness of the device is fixed, an empirical value can be set. Between the left and right boundaries, if the thickness value at a certain location is close to or exceeds the height of twice the thickness, it is indicated that overlapping of multiple devices has occurred. At this time, a warning is required.
c) Gap warning
Larger particles of dust may be present on the glass disc and if it happens that the device is on the dust at this time, a gap will appear between the device and the glass disc. Since the precision required for industrial inspection is in the order of microns, such gaps are sufficient to cause inaccurate thickness inspection of the device. And therefore must be warned.
The presence of a gap is detected using the following method:
1c) device region contour extraction
During the initial positioning of the device boundaries, the approximate locations of the left and right boundaries of the device have been obtained. And intercepting the binarized sub-image corresponding to the device according to the position, and then extracting the outline. The subsequent step is to detect the gap on the basis of the contour.
2c) Filtering of top and left and right profiles of devices
And the coordinates of the lower left corner and the lower right corner of the device can be approximately obtained by utilizing the bearing interface coordinates obtained in the initial positioning process and the coordinates of the left boundary and the right boundary of the device. Then, according to the two coordinate values, the top contour, the left side contour and the right side contour of the device are filtered out, and only contour points of possible gap areas are left.
3c) Gap region size measurement
The straight line segment remaining in the two steps can be regarded as the contour corresponding to the gap between the device and the glass plane. The contour points at this time are analyzed, and the boundary of the region is calculated. If the height of the region is less than 5 pixels and the width is greater than 20 pixels, then the region is considered to be a gap. At this point, a warning is required.
d) Filtering of dust interference
The dust has a great influence on the positioning of the sub-pixel boundary, and if the dust cannot be accurately filtered, the whole device thickness detection system cannot obtain a correct measurement result. Especially in China today, air pollution is serious, the specific gravity of particulate matters in air is high, and dust interferes with an industrial detection system very generally.
The invention performs dust filtration according to the following strategy:
1d) and performing local straight line fitting on the boundary coordinates of the sub-pixels: the boundary points are divided into a plurality of handwriting combinations tentatively by adopting the modes of edge growth and edge search:
first, the most coordinate contour points are selected and added to the set. Then, new adjacency points on the right are added to the set in succession. Every time a point is added, a straight line is fitted and then the distances to the straight line are calculated for all points in the set. If the average distance is greater than the threshold (1 pixel), the points are stopped from being added and the straight line sub-segment fitting is complete. Subsequent contour points are added to the new set.
2d) Selecting coordinate points with the distance larger than a threshold value;
and in each subset, selecting the coordinate points with the straight-line distance from the contour point to be greater than a threshold value.
3d) Segmenting the selected coordinate points;
and analyzing the coordinate points selected from the subsets, and dividing the coordinates into a plurality of subsections according to whether the coordinates are connected or not.
4d) The width of each subsection is inspected, and the width smaller than a threshold value is counted as dust;
5d) and recalculating the coordinate points, filtering the sub-segments corresponding to the dust, and replacing the coordinate points on the sub-segments with the coordinate points on the corresponding fitted straight lines.
e) Early warning of tailing phenomena
The placing position of the device on the glass disc is influenced by mechanical equipment, and the direction of the device can not be ensured to be vertical to the direction of the optical axis of the camera every time. When a large included angle exists between the device direction and the optical axis direction of the camera, the focal length of the camera is small, the device far away from the tail end of the camera cannot be focused correctly, and a gray area with high brightness appears at the tail end of the device image. This region directly causes problems in the thickness measurement of the device, and therefore this must be detected and pre-warned.
The following process is adopted to realize early warning of the trailing phenomenon:
1e) judgment of consistency of vertical coordinates of top boundary of device
When the trailing phenomenon occurs, the top boundary coordinates of the device are lower than the vertical positions of other top boundary coordinates on the device because the far end of the device occupies fewer pixels in the vertical direction in the image. The characteristic can be utilized to carry out preliminary early warning on the trailing phenomenon. The method comprises the following steps:
at the left and right ends of the device, a region of length 20 was selected each. Then, at the center position of the device, a region having a length of 50 is selected, and the vertical coordinates of the top contour points of the devices in the three regions are averaged, respectively.
If one area exists in the left end and the right end, the average coordinate of the area is compared with the average coordinate of the central area, and the difference value is larger than 5, the tailing phenomenon is considered to exist, and early warning is carried out.
2e) Analyzing brightness consistency of central region and head and tail end regions of device
As above, at the left and right ends of the device, a region of length 20 was selected, respectively. Then, at the center of the device, a region having a length of 50 was selected, and the average luminance of the three regions was calculated.
If one area exists in the left end and the right end, the difference between the brightness of the area and the average brightness of the central area is more than 20, the tailing phenomenon is considered to exist, and early warning is carried out.
Claims (7)
1. A fast and stable method for detecting the thickness of a sub-pixel-level precision device is characterized by comprising the following steps:
1) preliminary boundary positioning: dividing the image into a target area and an air area, and then obtaining a boundary of the two areas as a primary boundary positioning result;
2) and (3) sub-pixel accurate positioning: solving by adopting an arc tangent curve fitting method and a particle swarm optimization algorithm, and fitting by adopting a Bezier curve according to the positioning result of each column;
3) processing an abnormal condition;
wherein, the deviation between the initial positioning position of the preliminary boundary positioning and the final edge coordinate of the sub-pixel level is within the distance of 1 pixel;
the preliminary boundary positioning comprises the following steps:
a) coarse to fine boundary preliminary positioning
Obtaining a preliminary positioning coordinate point of the interface according to the following process:
1a) image reduction: firstly, reducing the image to one fourth of the original size;
2a) extracting a target area reference brightness value and an air area reference brightness value: pre-calculating brightness reference values of a target area and an air area in an image, dividing all pixels into two large classes according to brightness values by adopting a clustering method, and then calculating the brightness average value of each class to be respectively used as the reference values of the target brightness and the air area brightness;
3a) and (3) binarization operation: after reference brightness values of a target area and an air area are obtained, the mean value of the reference brightness values and the air area is used as a binarization threshold value of the whole image, and binarization operation is carried out on all pixels;
4a) extracting a maximum connected region: extracting a maximum connected region in a binarization operation result;
5a) preliminary boundary localization based on symmetry: obtaining a preliminary judgment by utilizing the symmetry of a pixel brightness change rule curve near the boundary;
6a) exception handling during initial positioning;
b) location of load bearing interface
1b) Setting the range of the inclination angle, traversing the angle, and horizontally projecting the binary image according to the set angle value to obtain a projection curve;
2b) analyzing projection curves under various angles, counting the number of lines required to be experienced when the projection value is increased from 0 to an image width value N, and selecting an angle value corresponding to the condition of the least number of lines as an inclination angle of a bearing interface;
3b) finding a projection curve corresponding to the inclination angle, carrying out difference calculation on the projection curve, calculating the difference between the upper and lower projection values, and selecting the line with the maximum difference value as the position of the bearing interface;
4b) then, filtering out a value corresponding to the bearing interface in the differential curve, and searching the maximum differential value to be used as an initial positioning position of the top boundary of the device;
c) positioning of left and right boundaries of the device:
1c) obtaining a boundary line between the top boundary position of the device and the bearing interface position by using the top boundary position of the device obtained in the previous step and the bearing interface position;
2c) extracting points of which the coordinate values are positioned above the boundary line from all the boundary points;
3c) according to the extracted points, two points on the leftmost side and the rightmost side are found;
4c) vertically projecting the binary image within 20 pixels around the leftmost point;
5c) differentiating the projection curve, and finding out the position with the maximum differential value as the left boundary of the device;
6c) within the range of 20 pixels around the rightmost point, vertically projecting the binary image, differentiating the projection curve, and finding out the position with the maximum differential value to obtain the right boundary of the device;
d) positioning of device top coordinate fillet
1d) Extracting the sub-outline corresponding to the device by using the bearing surface coordinate, the top coordinate of the device and the left and right boundary positions of the device obtained in the previous step;
2d) calculating a convex hull corresponding to the device sub-outline;
3d) generating a new binary image by using the convex hull, wherein the convex hull area is filled to be white;
4d) the difference is made between the image of the convex hull region and the image before the filling of the convex hull region is not carried out;
5d) in the difference map, four connected regions with the largest area are positioned;
6d) and calculating the positions of the four communication areas to obtain the fillet positions of the device.
2. The method according to claim 1, wherein the sub-pixel accurate positioning is achieved by fitting an arctangent curve to a column pixel curve near the boundary position, and the mathematical expression of the arctangent curve is as follows: y p1 atan (p2 x + p3) + p4, an accurate mathematical model of the luminance curve of the current column is obtained, then using the following formula: x0 ═ p3/p2, and the sub-pixel coordinates of each column were calculated.
3. The method according to claim 2, wherein the sub-pixel accurate positioning comprises the following steps:
a) direct estimation of two parameters, p1 and p 4: two parameters, p1 and p4, were calculated in advance using the shape of the luminance curve, and these two calculated parameter values were used as initial values for p1 and p 4:
the method comprises the following steps:
1a) gradually adding the pixels of the current column into the set from the upper part of the initial boundary position, if the standard deviation of the brightness of the pixels in the set is less than 5, continuing to add new pixels, if the standard deviation of the brightness of the pixels in the set is more than 5, stopping adding new pixels, and calculating the average value of the brightness of the pixels in the set at the moment as the average brightness of the upper side;
2a) gradually adding pixels of a current column into the set from the lower part of the initial boundary position, if the standard deviation of the brightness of the pixels in the set is less than 5, continuing to add new pixels, if the standard deviation of the brightness of the pixels in the set is more than 5, stopping adding new pixels, calculating the average value of the brightness of the pixels in the set at the moment to obtain the average brightness of the lower side, wherein the difference value between the average brightness of the upper side and the average brightness of the lower side is p 1;
3a) calculation of p 4: calculating the average value of all pixel brightness values on the brightness curve to obtain the vertical coordinate of the center point position of the boundary curve;
b) solving an arctangent curve based on a particle swarm algorithm:
the method comprises the following steps:
1b) setting a group of particles, 20 in total, each particle containing 4 parameters, and randomly setting an initial value to represent a solution of an arctangent curve;
2b) calculating the error between the curve corresponding to the current 20 particles and the actual brightness curve, and negating the error to serve as fitness;
3b) finding the most suitable curve parameter at present, then adjusting the direction and position of each particle, and continuing to calculate;
4b) repeating the steps 2b) and 3b) until the error of the best particle is less than the threshold value 10;
5b) calculating the accurate sub-pixel boundary position according to the formula x0 ═ p3/p 2;
c) application of the results of the previous column, initial values for p2 and p3 were obtained:
after the curve fitting for a column is finished, taking the values of p2 and p3 in the parameters of the column as the parameters of the brightness curve of the next column, and estimating p1 and p4 by calculation;
d) and (3) solving noise interference:
determining whether the fitting result of the current column is correct or not by judging the error in the fitting process, wherein for the situation of larger error, the column does not carry out accurate positioning of the boundary, and then the boundary is repaired by smoothing;
e) and (3) boundary smoothing:
and fitting the coordinates of the boundary points by adopting a Bezier curve fitting method.
4. The method according to claim 1, wherein the abnormal condition processing comprises
a) No device warning: detecting by analyzing a vertical coordinate on a glass interface, fitting the coordinate of the interface into a straight line, and then calculating the distance from each point to the straight line, wherein if the distance from all the points to the straight line is less than 3 pixels, no device exists, and early warning is needed when no device exists in an image;
b) early warning of a plurality of devices: the method comprises the following steps:
1b) the plurality of devices are separated from each other: inspecting the rising times and the falling times of the boundary coordinates, and if the rising times and the falling times exceed 1 time, early warning;
2b) multiple pieces of devices overlap: after left and right boundaries of the device are obtained by utilizing the ascending and descending rules of the device coordinate, the thickness distribution of the device between the left and right boundaries is considered, and if the thickness value of a certain position between the left and right boundaries is equal to or more than the height of twice the thickness, the overlapping of a plurality of devices is indicated to be generated, and early warning is carried out;
c) gap early warning:
1c) contour extraction of the device region: in the preliminary positioning process of the device boundary, the positions of the left and right boundaries of the device are obtained, a binarization subimage corresponding to the device is intercepted according to the positions, then the outline is extracted, and the subsequent steps are to detect the gap on the basis of the outline;
2c) and (3) filtering the contours of the top and the left and right sides of the device: obtaining coordinates of the lower left corner and the lower right corner of the device by using the coordinates of the bearing interface obtained in the initial positioning process and the coordinates of the left and right boundaries of the device, and then filtering the top contour, the left side contour and the right side contour of the device according to the two coordinates to only leave contour points of a gap area;
3c) gap area size measurement: after the processing of the two steps 1c) and 2c), the remained straight line segment is the profile corresponding to the gap between the device and the glass plane, the profile point at the moment is analyzed, the boundary of the region is calculated, if the height of the region is less than 5 pixels and the width is more than 20 pixels, the region is the gap, and early warning is needed;
d) filtration of dust interference:
1d) and performing local straight line fitting on the boundary coordinates of the sub-pixels: the boundary points are divided into a plurality of subsets in a tentative mode by adopting a mode of edge growth and edge search;
2d) selecting the coordinate points with the distance greater than the threshold value: in each subset, selecting out coordinate points with the linear distance from the contour point to be greater than a threshold value;
3d) and aiming at the selected coordinate points, carrying out segmentation: analyzing the coordinate points selected from the subsets, and dividing the coordinate points into a plurality of subsections according to whether the coordinate points are connected or not;
4d) the width of each subsection is inspected, and the width smaller than a threshold value is counted as dust;
5d) recalculating the coordinate points: filtering the sub-section corresponding to the dust, and replacing all coordinate points on the sub-section with coordinate points on a corresponding fitted straight line;
e) early warning of tailing phenomena
1e) And (3) judging the consistency of the vertical coordinates of the top boundary of the device: respectively selecting a region with the length of 20 at the left end and the right end of the device, then selecting a region with the length of 50 at the central position of the device, respectively calculating the average value of the vertical coordinates of the top contour points of the devices in the three regions, if one region exists in the left end and the right end, the average coordinate of the region is compared with the average coordinate of the central region, and the difference value is more than 5, then tailing phenomenon exists, and early warning is performed;
2e) and (3) analyzing the brightness consistency of the central region of the device and the regions at the head end and the tail end: as with 1e), selecting a region with the length of 20 at the left end and the right end of the device respectively, then selecting a region with the length of 50 at the center position of the device, calculating the average brightness of the three regions respectively, and if one region exists in the left end and the right end, the brightness of the region is different from the average brightness of the center region by more than 20, then tailing phenomenon exists, and early warning is carried out.
5. The method for detecting the thickness of a fast and stable sub-pixel level precision device according to claim 1, wherein in the step 5a), a column-wise search window with the length of 21 is set by taking a binarization boundary as a central point; then, the 21 pixels are inspected one by one, and a row of pixels with the length of 11 are collected on each pixel; then, the central symmetry of the 11 pixels is analyzed; and selecting the position with the best symmetry as an initial positioning boundary.
6. The method according to claim 1, wherein the step 6a) is a normal case without interference if the brightness variation of the pixels in the column is normal; if the brightness of the pixel of the column jumps many times, the interference condition is present.
7. The method as claimed in claim 4, wherein the step 1d) of heuristically dividing the boundary points into a plurality of subsets: firstly, selecting the contour point positioned at the leftmost side, and adding the subset; and then, adding new adjacent points on the right into the subset in sequence, fitting a straight line once every time when adding one point, then calculating the distance from all the points in the subset to the straight line, stopping adding the points if the average distance is greater than a threshold value, completing the fitting of the subsegment of the straight line, and adding the subsequent contour points into the new subset.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710295849.XA CN107218894B (en) | 2017-04-28 | 2017-04-28 | Rapid and stable sub-pixel precision device thickness detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710295849.XA CN107218894B (en) | 2017-04-28 | 2017-04-28 | Rapid and stable sub-pixel precision device thickness detection method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107218894A CN107218894A (en) | 2017-09-29 |
| CN107218894B true CN107218894B (en) | 2020-02-18 |
Family
ID=59943866
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710295849.XA Expired - Fee Related CN107218894B (en) | 2017-04-28 | 2017-04-28 | Rapid and stable sub-pixel precision device thickness detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107218894B (en) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108154506B (en) * | 2017-12-27 | 2020-07-03 | 合肥市雅视智能科技有限公司 | Automatic hexagonal nut detection method |
| CN109115127B (en) * | 2018-08-24 | 2020-10-30 | 合肥工业大学 | Sub-pixel peak point extraction algorithm based on Bezier curve |
| CN111242966A (en) * | 2020-01-15 | 2020-06-05 | 北京大豪科技股份有限公司 | Image boundary correction method, device, electronic equipment and storage medium |
| WO2021184175A1 (en) * | 2020-03-17 | 2021-09-23 | Yangtze Memory Technologies Co., Ltd. | Methods and systems for semiconductor structure thickness measurement |
| CN114687079B (en) * | 2020-12-31 | 2023-05-05 | 杰克科技股份有限公司 | System and method for managing equipment parameters of sewing equipment and computer storage medium |
| CN112818486B (en) * | 2021-02-02 | 2022-08-23 | 上汽大众汽车有限公司 | Hole position arrangement method for loudspeaker sound outlet holes |
| CN113536978B (en) * | 2021-06-28 | 2023-08-18 | 杭州电子科技大学 | Camouflage target detection method based on saliency |
| CN113804115B (en) * | 2021-08-10 | 2022-08-30 | 天津大学 | Concrete pouring base layer thickness monitoring method based on clustering |
| CN115841497B (en) * | 2023-02-21 | 2023-04-18 | 广州鲁邦通物联网科技股份有限公司 | Boundary detection method and escalator area intrusion detection method and system |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP5320216B2 (en) * | 2009-08-26 | 2013-10-23 | パナソニック株式会社 | Image processing apparatus, image processing system, and image processing method |
| CN104268857A (en) * | 2014-09-16 | 2015-01-07 | 湖南大学 | Rapid sub pixel edge detection and locating method based on machine vision |
| CN104359403A (en) * | 2014-11-21 | 2015-02-18 | 天津工业大学 | Plane part size measurement method based on sub-pixel edge algorithm |
| CN104573698A (en) * | 2014-12-31 | 2015-04-29 | 河海大学 | Image texture main direction recognition method oriented to imaging measurement |
| JP2015106095A (en) * | 2013-11-29 | 2015-06-08 | 三星ディスプレイ株式會社Ｓａｍｓｕｎｇ Ｄｉｓｐｌａｙ Ｃｏ．，Ｌｔｄ． | Image processing apparatus, image processing method, and computer program |
| CN104715487A (en) * | 2015-04-01 | 2015-06-17 | 大连理工大学 | Method for sub-pixel edge detection based on pseudo Zernike moments |
-
2017
- 2017-04-28 CN CN201710295849.XA patent/CN107218894B/en not_active Expired - Fee Related
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP5320216B2 (en) * | 2009-08-26 | 2013-10-23 | パナソニック株式会社 | Image processing apparatus, image processing system, and image processing method |
| JP2015106095A (en) * | 2013-11-29 | 2015-06-08 | 三星ディスプレイ株式會社Ｓａｍｓｕｎｇ Ｄｉｓｐｌａｙ Ｃｏ．，Ｌｔｄ． | Image processing apparatus, image processing method, and computer program |
| CN104268857A (en) * | 2014-09-16 | 2015-01-07 | 湖南大学 | Rapid sub pixel edge detection and locating method based on machine vision |
| CN104359403A (en) * | 2014-11-21 | 2015-02-18 | 天津工业大学 | Plane part size measurement method based on sub-pixel edge algorithm |
| CN104573698A (en) * | 2014-12-31 | 2015-04-29 | 河海大学 | Image texture main direction recognition method oriented to imaging measurement |
| CN104715487A (en) * | 2015-04-01 | 2015-06-17 | 大连理工大学 | Method for sub-pixel edge detection based on pseudo Zernike moments |
Non-Patent Citations (2)
| Title |
|---|
| 一种亚像素精度的边缘检测方法;孙秋成等;《北京工业大学学报》;20091031;第35卷(第10期);第1332-1337页 * |
| 基于反正切函数拟合的亚像素边缘检测方法;孙维健;《中国优秀硕士学位论文全文数据库 信息科技辑》;20151215(第12期);第I138-812页：参见摘要、第2.1-2.2节、第4-5章 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107218894A (en) | 2017-09-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107218894B (en) | Rapid and stable sub-pixel precision device thickness detection method | |
| JP6598162B2 (en) | Visual identification method of multi-type BGA chip based on linear clustering | |
| CN102175700B (en) | Method for detecting welding seam segmentation and defects of digital X-ray images | |
| JP6618478B2 (en) | Automatic in-line inspection and measurement using projected images | |
| CN113688807B (en) | Self-adaptive defect detection method, device, recognition system and storage medium | |
| US6771807B2 (en) | Method and system for detecting defects on a printed circuit board | |
| CN110930390B (en) | Chip pin missing detection method based on semi-supervised deep learning | |
| CN107239742B (en) | Method for calculating scale value of instrument pointer | |
| CN107341802A (en) | It is a kind of based on curvature and the compound angular-point sub-pixel localization method of gray scale | |
| CN110443278B (en) | Method, device and equipment for detecting thickness abnormality of grid line of solar cell | |
| CN108985337A (en) | A kind of product surface scratch detection method based on picture depth study | |
| CN113052797B (en) | BGA solder ball three-dimensional detection method based on depth image processing | |
| CN115170804B (en) | Surface defect detection method, device, system and medium based on deep learning | |
| TWI765442B (en) | Method for defect level determination and computer readable storage medium thereof | |
| CN105354816B (en) | A kind of electronic units fix method and device | |
| CN115144399A (en) | Assembly quality detection method and device based on machine vision | |
| CN115060162A (en) | Chamfer dimension measuring method and device, electronic equipment and storage medium | |
| CN111354047B (en) | Computer vision-based camera module positioning method and system | |
| Hashmi et al. | Computer-vision based visual inspection and crack detection of railroad tracks | |
| CN111476792B (en) | Extraction method of strip steel image contour | |
| CN115597494B (en) | Precision detection method and system for prefabricated part preformed hole based on point cloud | |
| CN107767372B (en) | Chip pin online visual detection system and method for layered parallel computing | |
| CN113096090B (en) | End face gap visual measurement method with chamfer, device, equipment and storage medium | |
| CN113192061A (en) | LED package appearance detection image extraction method and device, electronic equipment and storage medium | |
| CN114842041A (en) | Embedded-based steel rail profile detection light bar extraction method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20180112 Address after: No. 10 Xiang Pu Road, Feixi County, Hefei City, Anhui Applicant after: Hefei Yai Intelligent Technology Co., Ltd. Address before: No. 10 Xiang Pu Road, Hefei City, Feixi, Feixi, Anhui Applicant before: Hefei JAS vision Intelligent Technology Co., Ltd. |
|
| TA01 | Transfer of patent application right | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20200218 Termination date: 20210428 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |