Image Segmentation By Introducing Super-pixel Feature

Traditional interactive image segmentation methods,including region-based segmentation method and boundary-based segmentation method,depended on the user input prior information,only use the local relationship between pixels to build a relationship model,which is easy to be sensitive to the initial seed/contour position,and lack of robustness to noise,resulting in the problem of under-segmentation.Super-pixels divide the image into several sub-regions by clustering the pixels with similar features to accelerate process of subsequent tasks.This paper aims to improve the segmentation performance of the interactive segmentation method by using superpixels,and some research results have been achieved:1.The basic principles and disadvantages of the region-based and boundary-based interactive segmentation method are summarized,including that the region-based segmentation model is sensitive to noise and initial seed position and the boundary-based segmentation model is sensitive to initial contour position and noise and is prone to fall into local optimal during contour evolution.Meanwhile,the concept and advantages of hyperpixel and the classic hyperpixel segmentation model are summarized.2.In order to address the problem that the segmentation model based on region is sensitive to the initial seed location and noise,which is easy to produce the problem of under-segmentation,super-pixels are used to instead of pixels to enhance the connection between the seed point and the remote pixel,combined with the multi-layer super-pixels/pixel model to retain the image details.At the same time,sparse decomposition is introduced to optimize the model to improve the robustness of the model to image noise.The experimental results show that compared with the existing segmentation methods,the proposed method can obtain better segmentation effect and is robust to gaussian noise and salt and pepper noise.3.For boundary-based methods,such as active contour model,the segmentation results are easily affected by the initial contour position,and the contour evolution is prone to fall into the problem of local optimum and lack of accuracy and noise robustness.In this paper,the superpixel image region information is firstly extracted to construct a signal pressure function to prevent the contour from falling into the local optimum in the evolution process.Secondly,an energy functional based on super-pixel and pixel cooperative constraint is constructed to compensate for the fact that super-pixel cannot retain local details.At the same time,the model uses super-pixel blocks to accelerate contour evolution.Finally,sparse decomposition is used to optimize the model to reduce the influence of local noise on active contour evolution.Super-pixel is used to improve the traditional pixel-based model in this paper.The comparison experiment shows that the proposed method improves the robustness of the initial seed/contour position,and the introduction of sparse decomposition improves the robustness of the model to noise and the accuracy of model segmentation compared with segmentation model of only depending pixel.Finally,the experiment shows that the proposed algorithm exist some limitations,which need to be further studied and improved in the future.