Graph Cuts Image Segmentation Algorithm Researching Based On LCells Superpixels

Image segmentation is a fundamental problem in image processing, and the foundationof the computer vision problems such as pattern recognition, target tracking, scene analysis,etc. It’s a key issue in many high-level and semantic image processing methods. The mainpurpose of image segmentation is to divide the whole image into several disjoint pieces whichare consist inside. On this basis, extracts the object be interested in from complex backgroundfor subsequent understanding or analysis.At the same time, as a powerful preprocessing step of image segmentation algorithms,Superpixel generation algorithms have got extensive attention and long term development.There have more and more related new papers been published in top journals and conferences.A Superpixel clusters similar pixels in adjacent space, and takes some kind of total property todenote every pixel’s value belong to it. Using Superpixels instead of pixels as the basic unit ofimage processing, can reduce the information redundancy by cutting down sample points orsimplifying topology structure. Then algorithm’s efficiency will be much advanced.This paper introduces the background and history of the novel concept, also details theprinciple and characteristics of Superpixel algorithms. On this basis, this article improvedVCells Superpixel generation algorithm.The advanced algorithm called LCells Superpixelalgorithm. As same as VCells, LCells also starts with a hexagon grid as initialization, thentakes Lab-xy5d normalized distance to take place of EWCVT distance. After the local pixelgradient information is joined, the speed and accuracy of the algorithm have been improved.With UC Berkeley segmentation image benchmark database, the validity and superiority ofthe algorithm is verified according to the experiment.Incorporated with advanced LCells Superpixel algorithm, this paper improved GraphCuts-based image segmentation algorithm. Except applying LCells Superpixel algorithm topreprocessing steps to reduce the number of nodes in topology, how to improve the weightscalculation formula, let them reflect the property of the Superpixels better, is also one of thekey to keeping the overall segmentation accuracy. This paper details the advancedLCells-based Graph Cuts image segmentation algorithm. According to the experiment results,there is little loss of precision and accuracy, while the improving of efficiency at the same time. It’s a good-experience interactive image segmentation algorithm.