Superpixel Segmentation With Entropy Rate And Geometric Constraints

The main subject of this research is one of popular segmentation topics currently,Superpixel segmentation. Superpixel segmentation is an important content of imageprocessing, also a difficult problem in area of Computer Vision (CV). Superpixelsegmentation technique has now been used in medical image diagnosis, the militaryprecision strike, digital image processing, target tracking, pattern recognition, virtualreality scene,3D reconstruction, positioning layout of the building, atmosphericimage analysis and other fields, but also has obtained significantly results.First of all, based on entropy rate and adding the geometric constraint, thisresearch proposes a novel method combining the geometric constraint and entropyrate. Combining with the accuracy of entropy rate, the regularity of geometric shapeand the uniformity of balance item, this method segments image into homogeneous,compact, regular and uniform superpixels. Secondly, have been showed in thefollowing three aspects. Thirdly, we complete the segmentation of superpixel usingcombinatorial optimization and lazy greedy algorithm, and eventually obtain verygood results. Finally, there is a very clear sense of the application, this method issuitable for dealing with high-noise pictures, and exceeds the other methods in effectand efficiency.As above the forth contribution showed, in our real life, the processing of imagesin the noise and chaos scene, is a thorny and needed issue. So, this technic can begreatly promoted to other prospects, such as Medical Science, Military, ComputerApplication and so on. What’s more, the theory of entropy rate and geometrics can beused in other aspects of Computer Vision. At last, this issue has many improvements,so that get better superpixel.