Detection of product surface defects by learnable transform filters

Detection of surface defects on industrial products by machine vision technology is one of the main research topics. Surface scratchs, texture deformations and color differences are common problems at the industrial products. In this paper, a new method named learnable transform filters (LTF) are employed to detect surface defects. On learning stage, the transform operator is obtained using defected and undefected surface samples. On test stage transform operator is performed to detect defected surfaces on the product. Quality control operation is then ended by scaling defect of the product. In this study, LTF has been tested by synthetic and real product images. The results show that LTF presents satisfactory outcomes due to its learnable properties.