The Improved Mean Shift Image Segmentation Algorithm

Image segmentation is a challenging problem in the domain of image processing,and it is the foundation and critical step of image analysis,feature extraction,pattern recognition,etc.Image segmentation quality is very important,and the favorable information extracted from effective segmentation can make the image understanding possible.Scholars at home and abroad are continuously exploring in image segmentation,and have proposed thousands of segmentation methods.But,these methods could not apply to all the images,and many require a priori knowledge.Mean shift algorithm is a traditional statistical iterative method,has simple principle and without prior knowledge,and can deal with gray images?color images?complex images and large image datasets.However,the algorithm requires iterative calculation of each pixel in the image,therefore segmentation computational cost is high for practical tasks.Bandwidth of iteration process is difficult to determine,a bandwidth does not apply to all images.This article study around the two aspects of Mean Shift algorithm,to improve the segmentation efficiency and effect.These improvements make the algorithm more conducive to real-time image segmentation and more suitable for practical application.The main research works include:(1)Put forward a fast mean shift image segmentation algorithm(FMS),aim to improve the problem about too much iterations numbers and low segmentation efficiency of traditional Mean Shift algorithm.FMS algorithm chooses a small amount of pixels as initial point to iterative calculation,and other pixels are merging to the existing classes according to the similarity between the pixel and the class centers.As a result,the proposed FMS method reduces the total iterations numbers of Mean Shift algorithm,and boosts the segmentation efficiency.FMS algorithm using Berkeley Segmentation dataset and Internet images to do the experiments.(2)Put forward an image segmentation algorithm based on Support Vector Machine and FMS(FMS-SVM),aim to improve the problem that inappropriate iteration bandwidth will influence the segmentation results in Mean Shift algorithm.Firstly,FMS-SVM algorithm finds the smallest subgraph which contains the target area,and the following operations will conduct on the subgraph.As a result,the proposed FMS-SVM method reduces reduce the number of pixels to participate in the segmentation,and boosts the segmentation efficiency.And then,the FMS model is used for pre-segment of image,then SVM classifier is applied to classify the selected training samples.Finally,mapping back the subgraph segmentation result to the original image.FMS-SVM algorithm using Berkeley Segmentation dataset to do the experiments.In this paper,FMS and FMS-SVM image segmentation algorithm effectively solve the problems of traditional mean shift algorithm:low segmentation efficiency and difficult to select the iteration bandwidth.And the obtained research results will be significant to image segmentation technology.