CN106372645A - Mobile phone housing complex texture background defect detection method - Google Patents

Abstract

The invention relates to a mobile phone housing complex texture background defect detection method, and provides a defect detection algorithm based on mobile phone housing complex texture for the features that the texture of the surface of a mobile phone housing is complex, defects are various and the defects are tiny and are hard to detect. The method is characterized by, to begin with, dividing the surface of the mobile phone housing into a plurality parts, and carrying out amplifying pretreatment; then, carrying out image main structure extraction on the amplified mobile phone housing images to weaken the influence of amplified texture on detection; then, carrying out edge detection through an adaptive edge detection Canny method; and finally, extracting defects after a series of median filter, mathematical morphological operation, 8 connected region and area screening processing. By carrying out image main structure extraction on the amplified mobile phone housing images, the influence of amplified texture on detection is weakened.

Description

A kind of defect inspection method being applied to phone housing complex texture background

Technical field

The present invention relates to image deflects identification field, it is applied to lacking of phone housing complex texture background particularly to a kind of Sunken detection method.

Background technology

The inventive method is related to the image procossing of computer vision, particularly this part of image recognition, puts from one The process of defective locations is extracted in big phone housing picture.

Method the most close has Song Di [1] et al. to propose one kind based on gabor for Cellphone Accessories cut with the present invention With the Cellphone Accessories scratch detection algorithm of texture suppression, gabor filtering is carried out to metal surface image, extracts the skeleton of cut Structure, is suppressed the texture of metal surface, more accurately extracts cut with hysteresis threshold using anisotropic texture suppressing method.

Citation:

[1] Song Di, Zhang Dongbo, Liu Xia. the Cellphone Accessories scratch detection [j] based on gabor and texture suppression. computer work Journey .2014,40 (9): 1-5.

[2]rudin,l.,osher,s.,and fatemi,e.1992.nonlinear total variation based noise removal algorithms.physica d:nonlinear phenomena 60,1-4,259–268.

[3]li xu,qiong yan,yang xia,jiaya jia.2012.structure extraction from texture via relative total variation.acm transactions on graphics,31(6),139: 1-10.

[4]xianghua xie.a review of recent advances in surface defect detection using texture analysis techniques[j].electronic letters on computer vision and image analysis,2008,7(3):1-22.

Content of the invention

Substantial amounts of algorithm is all to compare clearly in image in some background texture at present, by colouring information, edge inspection Survey, gabor filtering, wavelet transformation etc. extract the obvious defect of imaging surface.These methods exist greatest problem be Image background is very big to the interference effect of defect, if background texture is somewhat more complex, is just difficult to detect by defect or flase drop. For this problem, the present invention be amplified for phone housing after the image having complex texture, first from complex texture In extract the main structure of image, then carry out adaptive rim detection, finally by a series of filtering and Mathematical Morphology Learn arithmetic operation, solve this problem.

The present invention proposes a kind of defect inspection method based on phone housing complex texture, first mobile phone case surface is drawn It is divided into some, the pretreatment being amplified；Then the main structure carrying out image to the phone housing picture after amplifying is extracted, Weaken the impact to detection for the texture after amplification；To carry out rim detection further according to a kind of auto-adaptable image edge detection canny method；? Afterwards according to extraction defect after a series of medium filtering, mathematical morphological operation, 8 UNICOM regions and area Screening Treatment.

Further, described auto-adaptable image edge detection canny method comprises the following steps:

Use Gaussian filter smoothed image；

Calculate amplitude and the direction of gradient with the finite difference of single order local derviation；

To gradient magnitude application non-maxima suppression；

With the detection of double threshold algorithms and link edge.

Further, phone housing defect is divided into four kinds, is scratch, cut, defect, spot respectively.

Brief description

Fig. 1 is significance detection method FB(flow block)；

Fig. 2 a is mobile phone sample graph；

Fig. 2 b is the figure after the amplification of mobile phone sample；

Fig. 3 a is texture artwork；

Fig. 3 b is texture histamine result figure；

Fig. 4 is edge detection results figure；

Fig. 5 a is figure after medium filtering；

Fig. 5 b is figure after mathematical morphological operation；

Fig. 6 is final result figure；

Fig. 7 a is scratch figure；

Fig. 7 b is cut figure；

Fig. 7 c is defect figure；

Fig. 7 d is spot figure；

Fig. 8 a is scratch result figure；

Fig. 8 b is cut result figure；

Fig. 8 c is defect result figure；

Fig. 8 d is spot result figure.

Specific embodiment

Below in conjunction with accompanying drawing, the present invention is described in detail:

Fig. 1 is phone housing defect inspection method FB(flow block).Main inclusion zoning enlarged drawing, extraction image master Seven aspects such as structure, rim detection, medium filtering, mathematical morphology.

Image semantic classification: Image semantic classification is broadly divided into two parts: carries out region division, enlarged drawing to phone housing Picture.Carry out region division: the phone housing picture of shooting is carried out being divided into several regions by we according to row, column, in order under The enlarged drawing of a part is prepared.As shown in Figure 2 a.Enlarged drawing: for the feature such as tiny of phone housing defect, to it In one piece, using micro capture digit microscope, it is amplified, as shown in Figure 2 b.

Image texture suppress: image texture suppression be built upon under phone housing complex background use in order to Eliminate the interference that complex texture background is extracted to defect.Rudin [2] et al. proposed one and is based on total variation in 1992 (rof) method of model, has been widely used in image reconstruction, recovery, denoising etc., and rof model is as follows:

$e_{tv-l2}=\min \left({\&integral;}_{\mathrm{\ω}}\right|\▿g|+\frac{\mathrm{\λ}}{2}||g-f|{|}_{l^2}^2)---\left(1\right)$

Wherein f and g represents input picture and output figure source respectively, and λ is a balance factor,Represent total variation,For calculating the l between f and g²Distance.

Li xu [3] et al. improves to rof model, and model is as follows:

Wherein i representing input images, p represents the index of 2d image pixel, behalf export structure image.

Q is the index of all of pixel in a square area centered on p point, and g is Gaussian function.

$g_{p,q}\&proportional;\exp (-\frac{{(x_p-x_q)}^2+{(y_p-y_q)}^2}{2{\mathrm{\δ}}^2})---\left(4\right)$

Contrast before and after result after texture suppression respectively as Fig. 3 a and Fig. 3 b.

Rim detection: canny edge detection operator its essence is makees smoothing operation with 1 quasi-Gaussian function, then to carry The first order differential operator positioning derivative maximum in direction, obtains Gaussian template derivative approximation according to variational method, in theory very The best edge operator being formed close to 4 linear combination of exponential functions, makees smoothing processing using Gaussian function to image, therefore has There is stronger noise removal capability.The concretely comprising the following steps of canny operator

Use Gaussian filter smoothed image；

Calculate amplitude and the direction of gradient with the finite difference of single order local derviation；

To gradient magnitude application non-maxima suppression；

With the detection of double threshold algorithms and link edge.

After the suppression of above-mentioned phone housing texture, as shown in Figure 4 with the edge detection results after the detection of canny operator.

Medium filtering and mathematical morphological operation: medium filtering sm (standard median filter) is that one kind has Less ill-defined non-linear filtering method, can not only remove or reduce random noise and impulse disturbances moreover it is possible to preferably Ground retains image edge information.This algorithm depends on fast reading sort algorithm, and its basic thought is in element to be sorted Arbitrarily choose an element in set and it is compared with other elements, all elements less than this element are all placed on Before it, all elements bigger than it are put after it；After a minor sort, can divide by the position that this element is located Boundary, set is divided into 2 parts；Then repeat said process to remaining 2 parts to be ranked up, until each section only remains Till next element；After the completion of all sequences, take the value of centrally located element in the set after sequence (i.e. so-called Intermediate value) as output valve.Traditional medium filtering can be defined as:

G (x, y)=med { f (x_i,y_j)}(i,j)∈m (5)

Wherein g (x, y) exports for medium filtering, f (x_i,y_j) for image pixel (x_i,y_j) gray value, m be template window Mouthful.First the information of defect can be become apparent from an expansive working before medium filtering, as shown in Figure 5 a.The form of image Student movement rice from the set theory development of mathematical morphology at last, let it be to the greatest extent, and elementary operation is very simple, but combinations thereof can To produce much complicated effect.It is exactly to move a structural element in the picture and carry out a kind of similar that morphological images are processed Operation in convolution.In each location of pixels, carry out one kind between structural element and the bianry image below it and specifically patrol Collect computing.The binary result of logical operationss preserves on the position corresponding to this pixel in the output image, and the effect of generation takes Certainly in the property of the size, content and logical operationss of this structure constitutive element.

(1) etching operation, its function is to carry out etching operation to image.Erosion operation is to eliminate all boundary points of object A kind of process, its result makes remaining object little several pixels than the original along its periphery.In image, little object is removed, rotten Erosion is very useful for removing little insignificant object in image.The definition of corrosion is:

$e=b\&circletimes;s=|x,y|s_{x,y}\&subsetequal;b|---\left(6\right)$

Wherein s be structural element that is to say, that by s to b corrosion produced by bianry image e be such point Ji Tai, If the origin translation of s will be completely contained in b to point (x, y) so s, the value of point (x, y) is exactly 1.

(2) expansive working, its function is to carry out expansive working to image.Expansion be corrosion inverse operations, be by with object All background dots of contact are merged into the process in this object.Result is that the border making object increases.Expand and filling up segmentation The cavity in object afterwards is very useful.The definition expanding is:

That is, s expands, to b, the Ji Tai that the bianry image d producing is made up of such point, if the initial point position of s So it is non-NULL with the common factor of b to move on to (x, y), and the value of point (x, y) is exactly 1.

The opening operation of image is first to corrode, to image, the process expanding afterwards.It has elimination small objects, at very thin point The inconspicuous effect changing its area during the border of separating objects and smooth larger object.Closing fortune is that image first expands post-etching fortune The result calculated.Closed operation has filtering function, can fill and lead up the internal ditch of image, hole and crack, so that broken string is connected.To binaryzation Image after process first carries out gray scale morphology closed operation and then carries out gray scale morphology opening operation again, the company to defect effectively Logical region is integrated, and can carry out effective de-noising to image after finally carrying out the medium filtering of 5*5 template again, and the most laggard The internal filling of row, effect is as shown in Figure 5 b.

8 UNICOM regions and area screening: after a series of filtering and morphology operations, in order to more easily right respectively Defect candidate region carries out feature decision, and we must be marked to each separate section in the result after processing first.This In employ 8 connection neighborhoods (neighborhood of pixel is eight points around pixel) to determine the label belonging to a certain pixel.For Although its size in the picture, shape are all uncertain for defect, but those excessively little regions can be specified and agree Fixed is not defective locations.The gross area of hypothesis image is area, area area (i) of its each candidate region, (wherein i=1, 2 ... n) when, when its area ratio n is less thanWhen (being set to 0.001 here), we just can consider that this block region is not Defective locations, it is more likely that being some less picture noises etc., thus can delete some empty according to area ratio False region, it is possible to achieve preferably image effect is processed, as shown in image 6.

The present invention has such advantages as with respect to prior art and effect (the invention compared with prior art, institute Have the advantage that and good effect.):

It is applied in the defects detection in image recognition, for best technology, often have ignored some complicated Background and tiny defect, so in some metal surfaces such as phone housing it cannot be guaranteed that its roughness can be up to standard, and the present invention In original technical foundation, first region division and amplification are carried out to phone housing, in some rim detection and filtering, Nogata Add texture to suppress it is ensured that some tiny defects can fully be detected in the base application of figure, complete good defect Identification.

The complexity of the size according to phone housing and superficial makings, the detection of roughness, carry out region to phone housing Division and amplification, phone housing defect after processing and amplifying is divided into four kinds, is scratch, cut, defect, spot respectively, situation Respectively as shown in Fig. 7 a-7d, the result finally detecting is respectively as Fig. 8 a-8d.

Above-described embodiment is the present invention preferably embodiment, but embodiments of the present invention are not subject to above-described embodiment Limit, other any spirit without departing from the present invention and the change made under principle, modification, replacement, combine, simplify, All should be equivalent substitute mode, be included within protection scope of the present invention.

Claims (3)

1. a kind of defect inspection method based on phone housing complex texture it is characterised in that: first mobile phone case surface is drawn It is divided into some, the pretreatment being amplified；Then the main structure carrying out image to the phone housing picture after amplifying is extracted, Weaken the impact to detection for the texture after amplification；To carry out rim detection further according to a kind of auto-adaptable image edge detection canny method；? Afterwards according to extraction defect after a series of medium filtering, mathematical morphological operation, 8 UNICOM regions and area Screening Treatment.

2. a kind of defect inspection method based on phone housing complex texture as claimed in claim 1 is it is characterised in that described Auto-adaptable image edge detection canny method comprises the following steps:

Use Gaussian filter smoothed image；

Calculate amplitude and the direction of gradient with the finite difference of single order local derviation；

To gradient magnitude application non-maxima suppression；

With the detection of double threshold algorithms and link edge.

3. a kind of defect inspection method based on phone housing complex texture as claimed in claim 1 is it is characterised in that mobile phone Shell defect is divided into four kinds, is scratch, cut, defect, spot respectively.