A Color Image Segmentation Algorithm Based On Super Pixel

Nowadays,the recording function is more and more powerful.Single word processing can not meet people’s needs,and it is the image processing requirements of increasingly stringent,so how can we achieve division works well,relatively complete details of extracting information,split time as much as possible has been a little hard research scholars wanted goal.This study is based on super-pixel color image segmentation algorithm.It were made two innovations from segmentation and split time complexity of these two areas.Faced with the basic operation in image processing unit pixels can not meet today’s efficient operation,we use super pixels instead of pixels as the subsequent processing of basic arithmetic unit,and reciprocal nearest neighbor proposed algorithm generates super pixel is given specific definition and nature of the reciprocal nearest neighbors,given the probability of reciprocal nearest neighbor structure analysis and illustrates how the algorithm by reciprocal nearest neighbor region merging.Then in the experiment all the pictures are from Berkeley gallery and datasets,to make a comparison with the algorithm runs in time,and with a few classic segmentation algorithms are made on the comparative experiments,the use of reciprocal nearest neighbor algorithm to generate super-pixel can either reduce the time complexity,but also get a good segmentation.The paper proposes to build the structural model based on weighted graph double super pixel set,and proposed a improved similarity measurement,through improved spectral clustering algorithm experiments in this paper after image segmentation can be obtained in a good segmentation results,while significantly reducing the segmentation error rate.