Research On Image Segmentation Algorithm Based On Superpixel

Image segmentation is a key step in image analysis and pattern recognition.Image segmentation is a technique that divides an image into areas of particular features and extracts the desired object from it.The quality of the image segmentation will directly affect the success of failure of subsequent image processing tasks,so the research of image segmentation technology is of great significance.Most of the traditional image segmentation methods use pixels as the processing unit,this method only considers the gray information and ignores the influence of spatial information between pixels,which makes the calculation efficiency lower when processing larger images.Therefore,in recent years,when scholars researched the field of image segmentation,they abandoned a large number of pixels that are computationally time-consuming.Instead,they focused on super-pixels.Superpixels refer to an area blocks that have similar pixels whose positions are adjacent and whose features such as color,brightness,texture are similar,Because superpixels can extract the local features of the image and obtain the local redundant information of the image,it can largely reduce the size of the object to be processed and the complexity of the subsequent processing,which is usually used as a preprocessing step of the image segmentation algorithm.This thesis studies and summarizes the current popular superpixel generation methods.Firstly,the image is initially segmented by using the superpixel method,and then the nearest neighbor merge algorithm and the maximum similarity merge algorithm(MSRM)are introduced to merge the superpixel regions,finally the segmentation result is obtained.The main contents of this paper is as follows:1.Firstly,the thesis summarizes the research background and research significance of the traditional image segmentation methods,and introduces the shortcomings of the traditional image segmentation techniques.The advantages and significance of the superpixel method are described.The basic idea and performance characteristics of superpixel segmentation are introduced.The Ncut algorithm,Scow algorithm and SLIC algorithm are mainly introduced.The six commonly superpixel methods are compared through experiments.Six typical algorithms are compared in view of segmentation performances according to the experimental results of four quantitative evaluations.The experimental results show that the SLIC algorithm can achieve the optimality in segmentation quality and accuracy compared with other algorithms.Therefore,this paper selects the SLIC algorithm as the preprocessing method for image segmentation.2.It is difficult to search the global optimal solution when using traditional region adjacency graph to describe the data structure.An image segmentation algorithm based on SLIC and fast nearest neighbor region merging has been proposed.First of all,the image is divided into small regions by SLIC superpixel algorithm,which the relation between regions is described by the adjacency table structure of the region adjacency graph(RAG)and the nearest neighbor graph(NNG).Then,the value of the dissimilarity function is calculated between each region to be merged with all of its adjacent regions.Finally,the region with the smallest similarity is merged.This method introduces the nearest neighbor graph to optimize the global search based on the region adjacency graph.This proposed algorithm can combine the most similar regions,and reduce the complexity of the merging calculation compared with the traditional region merging algorithm,which greatly improves the accuracy of the regional merging.3.For the requirement of real-time processing speed,an interactive image segmentation algorithm based on improved SLIC and dynamic region merging is proposed.Firstly,the original image is segmented by the improved SLIC superpixel algorithm.By manually demarcating part of the target and background area,and then using the MSRM algorithm to merge the similar regions,the algorithm in this section can be adaptively combine the most similar regions,compared to the merge algorithm that needs to set the threshold.Finally,experpiments show that the algorithm can achieves better segmentation results.