An Improved Segmentation Algorithm Based on Superpixel for Typical Industrial Applications

Recently, computer vision is widely used in the field of automatic production system, such as capsule detection, ceramic tile extraction, sole location, etc. However, such changes in lighting, texture of the background and colour of the material will have negative impacts on the performance of the system. In this paper, a robust object segmentation algorithm being able to accurately extract the target in different lighting, texture and material conditions is presented. First, automatic image enhancement techniques are applied to reduce the effects of illumination and to enhance the global differences between foreground and background objects. Second, image sharpening and edge-preserving smoothing techniques are used to enhance image edges and blur images, respectively. Smoothed images are then convoluted by gradient template to generate smooth gradient map. The sharpened image then subtracts the smoothed gradient image to produce the sharpened image with edge mask, which are then used as input to the super pixel algorithm. Third, the super pixels are clustered according to their spatial, colour and texture information, and the target is extracted by the region properties of the clusters. Experimental results show that the proposed algorithm has better segmentation accuracy or faster implementation time than Competitive algorithms.