Research And Application Of A Segmentation Algorithm For National Patterns' Basic Element

As treasure of national culture,national pattern carries the national spirit and cultural connotation,and reflects production level,economic level and aesthetic trend of each era.The basic element of the national pattern refers to the pattern that is independent and has certain cultural meaning in the national costume.Image segmentation means that one image is divided into several independent regions,and each region has a similar internal.This project comes from Beijing Municipal Science and Technology Commission.According to the characteristics of national pattern,the research on basic elements of national pattern is completed and the interactive segmentation system is realized.This paper studies kinds of segmentation algorithms and analysis their advantages and disadvantages.Based on the comparison of results of different segmentation algorithms on national patterns,this paper regards EGBIS algorithm as research emphasis and proposes an improved pattern segmentation algorithm SGB.Main work are as follows:(1)In view of the problem that EGBIS algorithm is inclined to over-segment,run SLICO as pre-segmentation to obtain homogeneous regions with high compactness and accurate boundaries.The homogeneous regions are described as super-pixels.Number of super-pixels is determined by the input pattern' size.(2)Regarding super-pixels as nodes to construct weighted undirected graph.Super-pixels are arranged irregularly.The number,location and size of adjacent nodes are unknown.In this paper,we define average region color as feature of a super-pixel,effectively reducing difference between adjacent nodes,and then complete the construction of edges using all adjacent super-pixels.On the basis of this special graph,define combination criteria to complete more desirable segmentation.In order to evaluate image segmentation algorithms,four indexes F-measure?Probabilistic Rand?Segmentation Covering and Variation of Information are introduced to compare EGBIS and SGB algorithms.The experiment results show that from subjective and objective aspects,SGB algorithm produces a relatively complete segmentation result,which is better than the original algorithm.(3)Based on project requirements and the above algorithm,an interactive national pattern segmentation system is set up in this paper.Under the environment of Studio Visual 2013 integrated with Qt5.3,the system includes four main functions:selection of interested regions,selection of algorithms and parameters setting,merging of sub-regions,display and storage of sub-regions.EGBIS and SGB algorithms are embedded in the system.Users can combine mean regions interactively based on the segmentation of algorithms,which can meet the needs of the target segmentation and get a more complete result.