Research Of Automatic Image And Video Segmentation Algorithm Based On Information Centroid

Nowadays,due to its practical value and its guiding role for theory development,image and video segmentation is still an important research direction in the field of computer vision and also the basic research content for many high-level vision applications that are based on image and video analysis,providing effective low-level features and research objects for them.However,what researchers are pursuing is the ideal automatic segmentation algorithms.But the efficiency and automaticity of the algorithms that are currently available is not ideal enough for all kinds of application scenarios.Therefore,this paper presents a set of solutions,using the information centroid as the core theory to guide the segmentation algorithm in efficiently and automatically segmenting images and videos.The research is as follows:First of all,this paper presents the information centroid algorithm.The information centroid theory is a clustering theory stating that there exists a fulcrum in an image to keep its balance and that a fulcrum has two properties,i.e.the information and coordinates of centroid.Based on this theory,this paper presents an abstract information centroid algorithm using the lever balance principle through mathematical deduction,which is then applied to subsequent image and video segmentation algorithm.Secondly,based on information centroid,this paper researches automatic image segmentation algorithm.Based on the idea that objects of interest can be easily segmented from an ideal saliency map,this paper adopts a process from saliency detection to segmentation.With pre-processing and SLIC superpixel segmentation,we get several approximately even regions,and each region is assigned a local information centroid.By clustering all local information centroids,a global information centroid is found.By comparing these local information centroids,the saliency map of uniqueness and distribution is generated.By using the global information centroid,a saliency map of uniqueness is intensified.After combining these two images,we find that the image is not smooth.We up-sample it.With the SaliencyCut algorithm,we automatically extract objects of interest.Based on information centroid,this paper researches automatic video segmentation algorithm.According to the information centroid-based saliency detection algorithm that we previously research,a saliency map is generated.For continuous frame images,the saliency map might not be ideal.Therefore,this paper presents a saliency defect detection algorithm based on information centroid.With it,we can come up with ideal saliency maps by detecting the rationality of Euclidean distances between the centroids of the saliency maps from continuous frame images.Based on high-quality saliency map,SaliencyCut algorithm often leads to over-segmentation.So this paper optimizes it and finally generates the ideal segmentation video.Finally,experiment results and analysis are provided in each chapter.For images,MSRA1 K dataset is used to make visualization and performance analysis for the saliency detection algorithm.For videos,DAVIS dataset is used to make visualization and performance analysis for the segmentation algorithm referred to in this paper.At last,the experimental results indicate that both algorithms have good robustness and prove the effectiveness of the information centroid algorithm.