Defect Detection Algorithm Based On Image Dictionary Representation

With the popularity of private car, security problem of cars has gotten more and more attentions. As the main components of a car, tire plays a vital role in the safety of cars. In recent years, competition in the tire industry is increasingly fierce. Product quality has become the key to get the core competitiveness of an enterprise. An effect tire defect detection method can ensure the quality of tire, which can reduce the rate of undetected defect tire.At present, the method of tire defect detection is mainly based on image processing on the X-ray image of tire internal structure. However, existing tire defect detection algorithms usually suffer from low accuracy, high complexity and low robustness, which are difficult to be used in production. Although image sparse representation method has a strong ability of comprehension, it has a relatively low efficiency.In the theoretical basis of the sparse representation, this paper proposed a defect detection algorithm based on fast K-SVD dictionary representation. First, a standard dictionary is obtained by training the defect-free images using fast K-SVD. Then the pseudo-inverse matrix of the standard dictionary is used to calculate the image representation coefficients in order to reduce the complexity. Finally, the defect is detected by analyzing the global and local features of image representation coefficients. The defect detection algorithm of fast K-SVD dictionary representation can effectively detect defects in the tire side. For the tire defects, which are contained in different types and positions of tire, they have features of small area and large variations in local grayscale. This paper further proposes an adaptive defect detection algorithm: fast PCA dictionary defect detection algorithm. This algorithm directly learns a dictionary from test image by fast PCA. Then the pseudo-inverse matrix of this dictionary is used to calculate the image representation coefficients fast. It can catch the defects by studying the reconstruction error. Because the PCA dictionary can match the test image adaptively, this algorithm can exactly detect the different defects in different positions in real time.Experiments of defect detection on large number of X-ray images show that the proposed defect detection algorithm based on fast K-SVD image representation can effectively detect defects in the tire side image. The defect detection algorithm based on fast PCA can effectively detect various types of tire defects in real time. Compared with the multi-scale analysis method based on wavelet transform and the method based on image component decomposition, fast PCA image representation defect detection algorithm has higher detection accuracy and efficiency, especially this algorithm is robust in different types of tire defects.