Research On The Algorithm Of Image Processing For Tire X-ray Detector

The quality and distribution of steel cords in the all-steel radial tire have a great impact on tire performance and service life. By using X-ray nondestructive detection, tire internal defects can be detected to reduce the production defect rate and decrease risks of tire in service. The computer vision technology can make nondestructive testing equipment fully automated and avoid the false due to various subjective factors in the process of artificial recognition.Firstly, this paper briefly introduces the composition of tire X-ray machine and presents the imaging system based on CMOS. The images collected by the imaging system are transmitted to the computer for analysis and processing. In the all-steel radial tire a variety of steel-cord problems may be caused during production, including broken cord, cross cord, impurity, bended cord, sparse cord, missing cord in a certain direction and so on. After comparing the features of these defects, they are divided into three categories: local flaws(impurity, cross and broken cord), curve defect(bended cord) and texture direction deviation(sparse cord and missing cord in a certain direction).The operations of image segmentation, enhancement and thinning are made during preprocessing. The tire image can be divided into belt and body region by using image segmentation algorithms based on Gabor function and C-means clustering. A reasonable Gabor filter is designed to separate the different texture energy, then clustering the filter results. Experiments show that this algorithm can effectively segment the tire image as required, comparing with the traditional threshold segmentation method and texture edge detection based on the gradient. To improve darker image and background interference, the gray-level correction algorithms are applied, and the results show that, compared with the histogram equalization and top-hat transform methods, the gamma correction of low gray extension can greatly improve the image contrast and remove the background. Image thinning of mathematical morphology is a precondition for the curve feature extraction.Two methods are proposed to identify the local defects. The first one is based on the edge feature extraction. After detecting the edges with Log operator, a 5×5 mask is proposed to recognize the local defects. The other method is used when the curve defect is detected. While tracking and loging all pixels in the refinement curve, the local defects are searched according to the current pixel and its 3-neighbor gray levels. Then judge bended cord from the slope-histogram, curvature-histogram and ratio between chord and its distance from arc. These two kinds of recognition algorithm can identify the related defects, but the first one is easily affected by the noise and it is limited to the application of a certain cord. The second one requires a better image thinning result. Lastly Log-Gabor filter is applied to detect texture direction defects in the belt region. Mean value and variance features are extracted from filter results in the directions of 30 degrees and 120 degrees to judge the defects from the feature distance.