Image Segmentation of Tiny Parts Based on Template Prior

In current industrial production, there is a problem of low precision and slow speed in manual measurement of tiny parts. In order to solve the problem of part size detection, we propose an accurate segmentation algorithm of tiny parts images based on template priors. In this paper, the detection object is the watch stator. The accuracy of watch stator is required to be micron level, and its size is within 25 mm. This method first realizes coordinate transformation on the CAD part template to achieve overall template matching, then uses local transformation to adjust the registration of local part edges. Finally, deep neural networks are used to implement image segmentation of parts by integrating edge shape prior. Combining the depth object segmentation with the geometric information of the part edge provided by the prior template, the edge structure of the part can be better restored for the blurred or damaged part image. The experimental results show that the image segmentation method based on the template prior can achieve high-efficiency and high-precision segmentation in tiny parts, and meet the requirements of automatic production line.