A novel method of sprite generation based on pixel similarity

In this paper, a novel, efficient, and robust method for sprite generation is presented. First a hierarchical block-based technique is applied to detect and segment the moving regions from the background. Next short-term and long-term motion estimation methods are combined to search the background motion relative to the generated sprite image. Considering the different attribution of pixels in different locations, pixel similarity (PS) is proposed to denote the contribution of pixels in sprite generation. Then a new blending technique based on pixel similarity is developed in order to prevent sprite image from being blurred due to the inaccurate foreground segmentation and global motion estimation. Experimental results show that the proposed method can significantly improve the sprite image quality.