Research On Image Segmentation And Moving Object Detection Based On Complex Network Theory

Image segmentation and object detection are always important research topics in the field of image analysis and computer vision.In recent decades,the theory of complex networks has developed rapidly,the application of complex network theory to the field of image analysis and computer vision has become one of the concerns of researchers.Based on the community detection theory and node degree distribution of complex network,the construction of image network,image segmentation and moving object detection algorithm in video are studied in this thesis.The specific research contents are as follows.(1)A new superpixel segmentation algorithm based on the modular increment of local network(LocalNet)is proposed.The adjacent weighted network of image is established by pixel color similarity.The neighborhood network is divided into communities by using the modular increment as the judgment criterion,and the superpixel segmentation of the image is completed.The LocalNet algorithm overcomes the shortcomings of the classic SLIC superpixel segmentation algorithm,which is easy to appear under segmentation.At the same time,superpixel segmentation algorithm is applied to over segment the image,and the superpixel region is used as the network node to construct the image network structure.Compared with the image network constructed by pixel nodes,the efficiency of image processing can be improved.The experimental results show that the LocalNet algorithm has higher segmentation precision than the superpixel segmentation algorithm SLIC,the improved algorithm VASLIC and MMTDSLIC.(2)An image segmentation algorithm based on local degree center theory(LDCS)is proposed.The image is segmented by using the LocalNet algorithm.The superpixel region is taken as the network node,and the color and texture similarity between regions are taken as the edge weights.The image adjacency weighted network structure is constructed.Combined with the local degree center,the local degree center of the image network is calculated,and the central node is taken as the center of the image network community to divide the network and complete the image segmentation.Experimental results show that the image segmentation accuracy of the LDCS algorithm is higher than that of the frame algorithm based on complex network theory(GFCNIS).(3)A moving object detection algorithm based on degree variation of image network nodes(IN-MOD)is proposed.The LocalNet algorithm is employed to segment each frame of the video.The adjacent weighted network structure of the image is established based on the regional color and texture similarity.The time series of the video network nodes is established according to the time sequence,and the degree change nodes of each frame of the image network node are found as candidate moving objects,In this thesis,LDCS image segmentation algorithm and K-means clustering algorithm are used to correct the detected candidate regions of moving objects.The experimental results show that the IN-MOD algorithm has a good detection effect when the camera shake and illumination change occur in the moving object detection.