复杂环境下的手部轮廓提取方法

摘要

To accurately segment the hand profile in complicated conditions, an improved watershed algorithm is pro-posed. Codebook is adopted to model the background for extracting the foreground. The skeletons of foreground and back-ground are extracted by thinning. These skeletons are used as markers in watershed transformation. Freeman chain code is introduced to smooth the hand profile in order to obtain one that is most close to visual effects. The sample picture has 1280 pixels × 720 pixels. The precision of segmentation results are assessed by two measures respectively based on dis-tance and region. The averaging deviation is within 5 pixels and misclassification error is within 9 percent. Experimental results demonstrate that the algorithm can effectively solve over segmentation and accurately extract the hand profile, which is also robust under complicated background or in varying illumination conditions.%为在复杂环境下准确分割出手部轮廓，提出了一种改进的分水岭算法。采用码本对背景建模以提取前景，提取出前景和背景的骨架，将骨架作为标记进行分水岭变换，利用Freeman链码平滑轮廓得到最贴近视觉效果的手部轮廓。样本图片为1280像素×720像素，从基于距离和基于区域两个测度来评价分割结果的精确度，平均绝对偏差在5像素以内，误分类误差在0.9%以内。实验结果表明，该算法能够有效解决分水岭的过分割问题，准确提取出多变的手部轮廓，对复杂背景和光照变化都有较好的鲁棒性。