CN106157303A - A kind of method based on machine vision to Surface testing - Google Patents

Abstract

The invention discloses a method for surface detection based on machine vision, which comprises the following steps: 1) adopting an LED ring light to directly collect images in the dark field frontal lighting mode; 2) using a dynamic threshold segmentation method to detect scratches in the surface area The region of interest is extracted; 3) Then, by using regional features and regional morphology, use corrosion operations on the surface area to remove noise points and small protrusions in the segmented area to ensure sufficient accuracy in the calculation; 4) Finally, use radiation transformation, Algorithms such as image smoothing and connected region extraction detect scratches and display the results. This method detects surface defects, which better reduces the interference of ambient light during the image acquisition process and separates metal defects in the most refined manner. Same as highlighted edges.

Description

A method of surface detection based on machine vision

technical field

The invention relates to the field, in particular to a method for surface detection based on machine vision.

Background technique

In the traditional product production process, the surface defect detection of products generally adopts the method of manual detection. With the continuous development of science and technology, especially the development of computer technology, computer vision detection technology has emerged. The system designed using this new technology is not affected by harsh environments and subjective factors, can quickly and accurately detect the quality of products, and complete detection tasks that cannot be completed manually. Machine vision inspection combines computer image processing and pattern recognition theory, and it integrates relevant knowledge in different fields such as computer technology, data structure, image processing, pattern recognition and software engineering.

In the process of detecting the metal surface, due to the limitations of the shooting environment, the scratches on the metal surface are highlighted in the black background area, and the metal edge is also highlighted. It is difficult to be detailed and accurate with the current technology Strict distinction is achieved through machine vision technology.

Contents of the invention

The technical problem to be solved by the present invention is to provide a method for surface inspection based on machine vision. This method detects surface defects, which better reduces the interference of ambient light during the image acquisition process; Metal defects and edges that are also highlighted are used to solve the above-mentioned multiple defects caused by the prior art.

In order to achieve the above object, the present invention provides the following technical solutions: a method for surface detection based on machine vision, comprising the following steps:

1) Use the LED ring light to collect images directly in the dark field frontal lighting method;

2) Use the dynamic threshold segmentation method to extract the region of interest for scratch detection in the surface area;

3) Then use regional features and regional morphology to use corrosion operations on the surface area to remove noise points and small protrusions in the segmented area to ensure sufficient accuracy in the calculation;

4) Finally, use algorithms such as radial transformation, image smoothing, and connected region extraction to detect scratches and display the results.

Preferably, in the step 2), the purpose of image thresholding is to divide the pixel set according to the gray level, and each subset obtained forms an area corresponding to the real scene, and each area has consistent attributes inside , while the adjacent region layout has this consistent property, the threshold segmentation operation is defined as

S={(r,c)∈Rgmin≤f _r , c≤gmax};

Therefore, threshold segmentation selects all the grayscale values in the image ROIR within a specified range of grayscale values into the output area S, so that gmin=0 or gmax=2 ^b -1, if the illumination can be kept constant, the thresholds gmin and gmax can be selected during system settings and never needs to be adjusted. Threshold segmentation is divided into fixed threshold segmentation and dynamic threshold segmentation;

Dynamic Threshold Segmentation The operation of comparing an image with its local background is called dynamic threshold segmentation processing. Let f _r , c represent the input image, and use g _r , c to represent the smoothed image. The dynamic threshold segmentation processing for bright objects is as follows

S={(r,c)∈Rf _r , cg _r , c≥g _diff };

The dynamic threshold segmentation process for dark objects is S={(r, c)∈Rf _r , cg _r , c≤-g _diff }.

Preferably, in the step 3), the algorithm adopted is a=R=∑(r, c)∈R1=∑ni-1cei-csi+1;

It can be seen from the above formula that the area a of the region is the number of points R in the region.

Preferably, in said step 4), the value of a point in the digital image or digital sequence is replaced by the median value of each point value in a domain of the point, so that the surrounding pixel values are close to the true value, thereby eliminating isolated noise point;

The method is to remove the two-dimensional sliding template of a certain structure, sort the pixels in the board according to the size of the pixel value, and generate a monotonically rising (or falling) two-dimensional data sequence. The result of threshold segmentation contains noise, which is not the last As a result, noise processing is suppressed by using image smoothing.

Preferably, in step 4), in the process of removing noise from the detection result, all connected regions with less than 4 pixels are regarded as noise and removed. In order to distinguish noise from defects, it is assumed that the noise is uniformly distributed, and Defects belonging to the same scratch are close to each other, therefore, small gaps in the defect area can be closed by dilation. In order to be able to calculate the connected area, it must be defined that two pixels should be considered as connected to each other.

The beneficial effect of adopting the above technical solution is: the method of the present invention detects surface defects, which better reduces the interference of ambient light during the image acquisition process, and most finely separates the metal defects and the same highlight the edge of. Not only that, the method is simple and easy to operate, and the feasibility is very strong.

Description of drawings

Fig. 1 is a control block diagram of the present invention.

detailed description

Preferred embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings.

Fig. 1 shows the specific embodiment of the present invention: a kind of method for surface detection based on machine vision comprises the following steps:

1) Use the LED ring light to collect images directly in the dark field frontal lighting method;

2) Use the dynamic threshold segmentation method to extract the region of interest for scratch detection in the surface area;

3) Then use regional features and regional morphology to use corrosion operations on the surface area to remove noise points and small protrusions in the segmented area to ensure sufficient accuracy in the calculation;

4) Finally, use algorithms such as radial transformation, image smoothing, and connected region extraction to detect scratches and display the results.

In the step 2), the purpose of image thresholding is to divide the pixel set according to the gray level, and each subset obtained forms an area corresponding to the real scene, and each area has consistent attributes inside, and the corresponding Neighborhood layouts have this consistent property, and the threshold segmentation operation is defined as

S={(r,c)∈Rgmin≤f _r , c≤gmax};

Therefore, threshold segmentation selects all the grayscale values in the image ROIR within a specified range of grayscale values into the output area S, so that gmin=0 or gmax=2 ^b -1, if the illumination can be kept constant, the thresholds gmin and gmax can be selected during system settings and never needs to be adjusted; threshold segmentation is divided into fixed threshold segmentation and dynamic threshold segmentation;

Dynamic Threshold Segmentation The operation of comparing an image with its local background is called dynamic threshold segmentation processing. Let f _r , c represent the input image, and use g _r , c to represent the smoothed image. The dynamic threshold segmentation processing for bright objects is as follows

S={(r,c)∈Rf _r , cg _r , c≥g _diff };

The dynamic threshold segmentation process for dark objects is S={(r, c)∈Rf _r , cg _r , c≤-g _diff }.

In the step 3), the algorithm adopted is a=R=∑(r, c)∈R1=∑ni-1cei-csi+1;

It can be seen from the above formula that the area a of the region is the number of points R in the region.

In said step 4), the value of one point in the digital image or digital sequence is replaced by the median value of each point value in a domain of the point, so that the surrounding pixel values are close to the true value, thereby eliminating isolated noise points;

The method is to remove the two-dimensional sliding template of a certain structure, sort the pixels in the board according to the size of the pixel value, and generate a monotonically rising (or falling) two-dimensional data sequence. The result of threshold segmentation contains noise, which is not the last As a result, noise processing is suppressed by using image smoothing.

In the step 4), in the process of removing noise from the detection result, all connected regions with less than 4 pixels are regarded as noise and removed. In order to distinguish noise from defects, it is assumed that the noise is uniformly distributed and belongs to the same division The defects of traces are close to each other, therefore, the small gaps in the defect area can be closed by dilation. In order to be able to calculate the connected area, it must be defined that the appropriate two pixels should be considered as connected to each other.

The method of the invention detects surface defects, and better reduces the interference limited by ambient light during the image acquisition process. The most refined result is the separation of metal defects and edges that are also highlighted. Not only that, the method is simple and easy to operate, and the feasibility is very strong.

The image is collected by using the LED ring light direct dark field frontal illumination method, and then the dynamic threshold segmentation method is used to extract the area of interest for scratch detection in the surface area, and then the corrosion operation is used on the surface area by using area features and area morphology Remove noise points and small protrusions in the segmented area to ensure sufficient calculation accuracy, and finally use algorithms such as radial transformation, image smoothing, and connected region extraction to detect scratches and display the results.

There are generally two types of directionality in lighting: diffuse and direct.

When diffused, the intensity of light is nearly the same in all directions. When directly illuminated, the light emitted by the light source is concentrated in a very narrow spatial range. The detection object in this paper is surface scratches. Due to the small detection area of such defects and the inconspicuous scratches, it is difficult to obtain ideal scratch images under bright field illumination. Therefore, this test adopts the direct dark field lighting method of LED ring light. The ring light has a very small angle with the surface of the object, which can highlight the gaps and protrusions of the measured object, so scratches, textures or engraved characters are enhanced. See more clearly.

The captured image does not provide information about the objects contained in the image. In order to obtain the object information in the image, image segmentation must be performed. Image segmentation is to divide the image into some areas. In the same area, the image features are similar; in different areas, the image features are quite different. Image features can be the features of the image itself, such as the grayscale of pixels, edge contours, and textures. Image thresholding segmentation is one of the most commonly used and simplest image segmentation methods. The purpose of image thresholding is to divide the pixel set according to the gray level, and each subset obtained forms an area corresponding to the real scene. Each area has consistent attributes inside, and the layout of adjacent areas has this consistency. Attributes. The threshold segmentation operation is defined as

S={(r,c)∈Rgmin≤fr, c≤gmax}

Therefore, threshold segmentation selects all the points in the image ROIR whose gray value is within a specified gray value range into the output area S. Let gmin=0 or gmax=2b-1. If the illumination can be kept constant, the thresholds gmin and gmax can be selected during system setup and never need to be adjusted. Threshold segmentation is divided into fixed threshold segmentation and dynamic threshold segmentation. Dynamic threshold segmentation The operation of comparing an image with its local background is called dynamic threshold segmentation processing. Use fr, c to represent the input image, and use gr, c to represent the smoothed image. The dynamic threshold segmentation processing of bright objects is as follows

S={(r,c)∈Rfr,c-gr,c≥gdif}

And the dynamic threshold segmentation processing for dark objects is

S={(r,c)∈Rfr,c-gr,c≤-gdiff}

In dynamic threshold segmentation, the size of the smoothing filter determines the size of the object that can be segmented. If the filter size is too small, the estimated local background at the center of the object will be suboptimal.

After the previous processing, the region extracted from the image or the sub-pixel precision contour can be obtained. But they only contain raw descriptions of the segmentation results. Later, certain regions or contours must be selected from the segmentation results to be removed as unwanted parts in the segmentation results. By far the simplest region feature is the area of the region:

a=R=∑(r,c)∈R1=∑ni-1cei-csi+1

It can be seen from the above formula that the area a of the region is the number of points R in the region. If the region is represented by a binary image, then use the first summation equation in the formula to calculate the area of the region; if the region is represented by run-length encoding, then use the second summation equation in formula 4 to calculate The area of the region. A region can be regarded as a union of all its strokes, and the area of each stroke is very easy to calculate. Note that the second accumulator has much fewer terms than the first accumulator. Therefore, the stroke representation of the region can make the calculation of the area much faster, and this feature is applicable to almost all regional features.

After the previous series of processing, defect detection can be performed on the region of interest, and the dynamic threshold segmentation operation needs to be used again to detect defects, and the median filter can be used to estimate the background.

The basic principle of median filtering is to replace the value of a point in a digital image or digital sequence with the median value of each point in a domain of the point, so that the surrounding pixel values are close to the real value, thereby eliminating isolated noise points. The method is to remove a certain structure of the two-dimensional sliding template, sort the pixels in the board according to the size of the pixel value, and generate a monotonically rising (or falling) two-dimensional data sequence. The result of threshold segmentation contains noise, which is not the final result. Noise processing, suppressed by using image smoothing.

Through the above operations, the detection of surface scratches basically ends, because in the process of removing noise, all connected regions with less than 4 pixels are regarded as noise and removed. In order to distinguish noise and defects, it is assumed that the noise is uniformly distributed, and the defects belonging to the same scratch are close to each other. Therefore, small gaps in the defect area can be closed by dilation. In order to be able to compute connected regions, it must be defined that two pixels should be considered connected to each other. The above is the detection process, and the desired detection result can be obtained through the above operations.

Continuously read in the template of the surface scratch image from the specified directory, and set the image size, and use the LED ring light to directly illuminate the surface scratch image in the dark field.

The scratches are highlighted in the black background area, but the edge of the surface and the edges of the four inner squares in the surface plane part are also highlighted. In order to distinguish the scratches from the edge of the surface, the bright edge area is segmented first. The segmented region is then subtracted from the region of the surface, thereby narrowing down the region of interest for scratch detection to the subtracted region.

Through the above processing, the next step is to determine the plane to be detected, so the region of interest needs to be extracted. It is necessary to remove the bright borders of the surface and the bright borders of the four small squares in the middle from the segmentation results. First of all, the direction and size of the surface in the image must be known. In order to obtain the direction and size of the surface, the 4 inner squares are segmented using the area morphology again. Firstly, two closing operations are used to fill the small holes on the edge of the previously divided internal square, and there are gaps on the boundary of the internal square.

So far, the scratches are still in the segmented bright border area. To be able to detect scratches, the scratches need to be separated from the segmentation results. Since the shape of the boundary region of the inner square is known, the scratches can be removed using a suitable structuring element opening operation. For this a structuring element is generated consisting of two rectangles with parallel axes representing the two opposite sides of the inner square.

Structural elements may not be rotated when generating rectangles in the proper orientation. But the center of the rectangle needs to be transformed according to the orientation.

The opening operation can be used as template matching, and will return all points in the input area that match the structure elements. As expected, the result contains an inner square boundary. However, the result still contains the outer boundary of the surface part, because the distance from the inner square to the surface boundary is the same as the side length of the inner square. In order to remove the boundary part, the intersection of the result of the open operation and the etched surface area is taken.

This results in the region RegionSquares with only 4 inner square boundaries. The last surface to check is the difference between the surface area and the boundary of the inner square.

The surface area is eroded using circular structuring elements to remove boundaries before computing the difference. The radius of the circle is the sum of Border-Width and BorderTolerance, both values are defined in advance. The radius plus BorderTolerance is to remove pixels that are very close to the border during detection. The grayscale of these pixels will be affected by the border and may be misjudged as a defect. In the same way, the boundary area of the representative inner square should also be inflated. The resulting region of interest Re-gionSurface containing the surface detection plane. Note that the surface white border and the inner square white border are not included in the region.

After the above processing, defect detection can now be performed on the region of interest: the dynamic threshold segmentation operation is used again to detect defects, and the median filter can be used to estimate the background at this time. Based on the known maximum scratch width ScratchWidthMax, use Scratch-WidthMax as the median filter radius to remove all scratches. Due to the darkfield frontal illumination, scratches appear as bright areas in the image, which can be easily segmented using the predefined ScratchGrayDiffMin as a threshold.

In this case, all connected regions with less than 4 pixels are considered as noise and removed. Not all the noise is completely removed, further increasing the threshold may remove some discontinuous defect regions at the same time. In order to distinguish noise and defects, it is assumed that the noise is uniformly distributed, and the defects belonging to the same scratch are close to each other. Therefore, small gaps in the defect area can be closed by dilation. The original disconnected defects are connected together after expansion, and the connected area is recalculated for the expanded area. In order to get the original shape of the defect, the intersection of the original area before expansion and the connected area is taken.

Note that the intersection operation does not affect the connectivity of each component, so only the outline of the connected region is increased by dilation. Finally, all regions larger than the predetermined minimum scratches are selected.

The above is only the preferred embodiment of the present invention, it should be pointed out that for those skilled in the art, without departing from the inventive concept of the present invention, some modifications and improvements can also be made, and these all belong to the present invention. protection scope of the invention.

Claims (5)

1. one kind based on the machine vision method to Surface testing, it is characterised in that comprise the following steps:

1) LED circular lamp direct dark field frontlighting mode is used to gather image；

2) dynamic threshold segmentation method is used by the region of interesting extraction of scratch detection in region, surface out；

3) again by use provincial characteristics, regional morphology region, surface use erosion operation is removed in cut zone miscellaneous point and Little outthrust, it is ensured that have enough precision in calculating；

4) radiation conversion, image smoothing, connected region is finally used to extract scheduling algorithm and detect cut and show its result.

Method based on machine vision to Surface testing the most according to claim 1, it is characterised in that described step 2) In, the purpose of image threshold is according to gray level, and collection of pixels carries out a division, and each subset obtained forms one The region corresponding with real-world scene, has consistent attribute, and adjacent area layout has this consistent genus inside regional Property, Threshold segmentation operation is defined as

S={ (r, c) ∈ Rgmin≤f_r, c≤gmax}；

Therefore, in gray value in image ROIR is in a certain appointment intensity value ranges by Threshold segmentation, all point chooses output area In S, make gmin=0 or gmax=2^b-1, if illumination can keep constant, threshold value gmin and gmax can be determined when system is arranged Select and never with being adjusted；Threshold segmentation is divided into fixed threshold segmentation and dynamic threshold segmentation；

The operation that image and its local background are compared by dynamic threshold segmentation is referred to as dynamic threshold segmentation and processes, and uses f_r, c Represent input picture, use g_r, c represents the image after smoothing, then processes as follows to the dynamic threshold segmentation of bright object

S={ (r, c) ∈ Rf_r, c-g_r, c >=g_diff}；

And the dynamic threshold segmentation process to dark object is S={ (r, c) ∈ Rf_r, c-g_r, c≤-g_diff}。

Method based on machine vision to Surface testing the most according to claim 1, it is characterised in that described step 3) In, the algorithm of employing is a=R=∑ (r, c) ∈ R1=∑ ni-1cei-csi+1；

From above formula, the area a in region is exactly the R that counts in region.

Method based on machine vision to Surface testing the most according to claim 1, it is characterised in that described step 4) In, it is to carry the Mesophyticum of each point value in territory one of this point of the value of any in digital picture or Serial No. to replace, allows around Pixel value close to actual value, thus eliminate isolated noise spot；

Method is the two-dimentional sleiding form of certain structure, pixel in plate is ranked up according to the size of pixel value, generates single Adjust rise (or decline) for 2-D data sequence, containing noise in the result of Threshold segmentation, this is not end product, noise Process, by use image smoothing suppress.

Method based on machine vision to Surface testing the most according to claim 1, it is characterised in that described step 4) In, testing result is during removing noise, and all connected regions being less than 4 pixels are counted as noise and are removed, for Differentiation noise and defect, it is assumed that noise is equally distributed, and the defect belonging to a cut together is close to each other, therefore, By expanding, gap medium and small for defect area can be closed；In order to calculate connected region, it is necessary to define suitable two Pixel should be considered to communicate with each other.