Research On Image Segmentation Method Based On Super-pixel

As a very active branch in the field of artificial intelligence,image understanding is an extension of traditional computer vision,and in order to better understand the image quickly and accurately,a variety of image segmentation methods emerge one after another.Image segmentation is generally used as a pre-processing step in the field of image understanding.By dividing the image into regions that do not overlap each other,the difficulty of image understanding is reduced.Aiming at the problems of conventional image segmentation or slow running or unsatisfactory segmentation effect,a simple and fast image segmentation method based on super-pixel is proposed in combination with image boundary information.The segmentation process is mainly divided into three parts: boundary information extraction,super-pixel generation and super-pixel fusion.In the boundary information extraction part,from the traditional methods and deep learning methods,each selects the boundary extraction algorithm with the best comprehensive performance to verify the influence of the boundary extraction algorithm on the segmentation result in the experiment;in the super-pixel generation part,according to the obtained boundary information improves the original super-pixel generation algorithm,improves the boundary consistency of generating super-pixels,and makes the segmentation result more accurate.In the super-pixel fusion part,the boundary information is first strengthened by calculating the chi-square distance of the regional color features.Then connect the two adjacent super-pixel blocks by the enhanced boundary information,and then a super-pixel is merged using a hierarchical merging algorithm based on greedy strategy,and finally the image segmentation is completed.The experimental results show that the proposed super-pixel generation algorithm with boundary information is better than the original super-pixel generation algorithm,and the improved algorithm can increase the recall rate of boundary by more than 5%,and the achievable segmentation accuracy is higher.Compared with the existing segmentation method,the based on super-pixel image segmentation method can increase the segmentation coverage by up to 4% and the segmentation speed by more than 10 seconds.