摩擦片表面缺陷的视觉显著性检测算法

针对摩擦片表面缺陷高精度高效率的检测要求以及摩擦片自身复杂的表面状况，提出基于视觉显著性的检测算法. 利用图像分割，将摩擦片从背景中分离；使用高斯平滑弱化表面纹理，采用多尺度细节增强算法补偿高斯平滑中丢失的缺陷边缘信息，计算图像中目标的显著性进行强弱分化；采用连通域法和OTSU，提取缺陷区域的二值图像. 经由实验验证，该算法针对摩擦片的缺陷检测具有较强的针对性，缺陷识别率超过98%，双面检测100个摩擦片用时27 s. 从客观和主观两个方面对检测结果进行评价验证，结果表明，该算法具有较高的识别率和精确度，满足工业检测的需求.

Visual salience detection algorithm for surface defects of friction sheets

A detection algorithm based on visual salience was proposed according to the requirement of high precision and efficiency for detecting the surface defect of friction sheet and complex surface conditions of the friction sheets themselves. The friction sheets were separated from background by the image segmentation. The surface texture was smoothed by Gaussian Blur. The multi-scale detail enhancement algorithm was used to compensate missing defect edge information in Gaussian Blur, and the saliency of the target in this image was calculated for differentiation. The connected domain method and Otsu were utilized to extract the binary images of the defect area. The experimental results show that the algorithm has strong pertinence for the defect detection of friction sheets. The defect recognition rate is over 98%. It takes 27 s to detect 100 friction sheets on both sides. From objective and subjective aspects, the detection results prove that the algorithm has high recognition rate and accuracy to meet the demand of industrial assembly.