Abstract: To improve the efficiency of leather defect detection, a leather defect detection algorithm is proposed by combining photometric stereo with saliency object detection. First, a acquisition platform for photometric stereo is established to collect leather samples in different orientation. Then, photometric stereo is applied to obtain a synthetic image and a surface normal vector image. Next, the surface normal vector image is processed by curvature filtering and the synthetic image or the filtered image is selected by the approximate surface roughness features. Finally, defect detection is performed based on saliency object detection. Experimental results indicate that, compared to the existing methods, the proposed algorithm can detect several kinds of defects on different leathers made of different materials at a high accuracy and a high speed.

Key words: leather defect detection, photometric stereo, curvature filtering, saliency object detection, image segmentation

摘要： 为了提高皮革缺陷检测效率，提出一种基于光度立体视觉和图像显著性的皮革缺陷检测算法。搭建光度立体视觉平台，完成不同角度的皮革样本采集，利用光度立体视觉技术计算皮革样本的合成图和表面法向量图；对表面法向量图进行曲率滤波操作，用近似表面粗糙度特征自适应选择合成图或滤波图；利用显著性目标检测算法完成皮革缺陷检测与定位。实验结果表明，与现有皮革缺陷检测方法相比，该算法能很好地检测不同材质皮革的多种缺陷，且准确率高，速度快。

关键词: 皮革缺陷检测, 光度立体视觉, 曲率滤波, 显著性目标检测, 图像分割