Scratch Detection Used For Mobile-phone Partition Based On Machine Vision

The surface scratch is one of the most frequently occurring defects ofmobile-phones. Due to the impact of poduct processing, the surface scratches appearas variable appearence. Moreover, the background texture is complicated, whichbrings a great difficulty for scratch extraction. With the continuous improvement ofproduct quality standards in international markets, manufacturers put forward higherrequirements for appearance quality of products. The scratch width standards of somephone maker already reached micron level. For the high-quality product inspection,manual detection has been completely unable to meet the requirements. Inspectionbased on machine vision is the important technology to detcect the scratches ofmobile-phone partition, and it is of great scientific and industrial value. Surfacescratches detection based on machine vision products have been researched andreported in the world, but no specifically work has been achieved for mobile-phonemetal partition. Considering the interference of metal textures of partiton, and thereal-time application of the machine vision inspection system, this paper mainlyfocused on the problem of mobile-phone partition surface scratches, and a machinevision detecting system is designed to achieve automatic scratch detection. The mainwork in this paper includes:(1) The design of machine vision detecting system of mobile-phone partition,which achieves automatic detection of the scratches, including program design ofimage detection system, detection protocols, hardware equipment, systems-processcontrol program, PLC control program and so on.(2) Combined anisotropic texture suppression algorithm with Gabor filters, amethod for mobile-phone metal partition scratches extraction is proposed in this paper.The method takes full advantage of the characteristics of Gabor filter function, toextract the skeleton structure of scratches. Then through anisotropic texturesuppression algorithm, the interference of the background are greatly suppressed.Finally, combined with hysteresis multi-threshold segmentation method, whichminimized background texture interference， the proposed method accurately extractthe scratches of mobile-phone partitions.(3) A scratch detection algorithm that combine linear operator and SVMclassification is proposed in this paper. By adding a vertical auxiliary linear operator,we improve the current linear operators to achieve feature extraction. Then a linear support vector machine (SVM) is applied to classify the pixels in the image.Compared with the former detection methods, detection algorithms in this paper haveseveral advantages: i.e., required less features, and fewer samples are needed fortraining.The surface scratches detection method designed in this paper mainly solvesbackground textures interference present in background. Compared with the previousmachine vision detection method, the method proposed in this paper has showed astronger robustness to noise. This algorithm is relatively easier to implement and hasreal application ability. The modular designed system has strong maintainability andapplicability.