Image Jigsaw Puzzles with A Self-Correcting Solver
Jigsaw puzzle is an intellectual game and serves as a platform for many scientific applications.Several computational methods have been proposed to deal with the jigsaw puzzle problem in recent years.However,there are still some drawbacks.First,these methods fail to consider the content consistency of the reconstructed images.Specially,the traditional measures only reflect similarity between adjoining pieces but neighboring pieces.Second,these methods cannot guarantee the overall reconstruction correctness,because the strategy of assembly merely tries to correct the measure of adjoining pieces at each step.To overcome these drawbacks,this paper proposes a new method which contributes the follows: 1) A new measure considers the transmission relationships of four neighboring pieces to make better use of content consistency.2) A self-correcting mechanism avoids error accumulation of adjoining matrix and improves the overall accuracy of assembly,which is achieved through ordering the pairwise relations.Experimental results on 20 images demonstrate that the proposed method significantly improves the performance and outperforms the state-of-the-art methods.
Jigsaw puzzle dissimilarity-based measure assembly self-correcting mechanism minimal spanning tree
Xiangtao Zheng Xiaoqiang Lu Yuan Yuan
Center for OPTical IMagery Analysis and Learning(OPTIMAL),State Key Laboratory of Transient Optics a Center for OPTical IMagery Analysis and Learning(OPTIMAL),State Key Laboratory of Transient Optics a
国际会议
西安
英文
112-118
2013-09-14（万方平台首次上网日期，不代表论文的发表时间）