ε-Isometry based shape approximation for image content representation

摘要: Shape approximation is usually a prerequisite step to image content analysis and understanding has been well studied in the passed decades. However, those approaches show their deficiencies while facing factors such as representation efficiency, variation of scale initial estimation. To alleviate these issues, we propose novel method for @e-isometry based shape approximation. We first analyze descending property on approximating error its relation with salient geometric features. After that, approximate polygonal detect feature point @e-isometric construction. In experiments, employ traditional benchmarks, MPEG7 dataset, SQUID dataset other real contours evaluate visual effects quantitative performances proposed method. Experimental results demonstrate that our not only robust estimation, but also outperforms state-of-the-art methods respect compactness variability.