去除冗余背景的手势连续轮廓提取算法与实验  

Algorithm and Experiment of Continuous Contour Extraction for Gesture without Redundant Background

作　　者：张钰[1] 徐伟斌 陆敬微 王光义[1] 

机构地区：[1]杭州电子科技大学电子信息学院国家级电工电子实验教学示范中心,杭州310018

出　　处：《实验室研究与探索》2017年第10期8-12,99,共6页Research and Exploration In Laboratory

基　　金：国家自然科学基金面上项目"CIS像素RTS噪声的时变物理统计建模与多维非均匀抑制方法研究"(61372156)

摘　　要：连续轮廓提取是三维图像模式识别和重建的重要步骤。现有轮廓提取方法处理后的轮廓存在断点,严重影响后续图像处理步骤的质量。提出了一种手势连续轮廓提取方法。采用高斯混合模型定位方法确定视频图像中手势所在区域;利用Sobel算子对所选择区域中手势进行粗糙轮廓提取;采用深度优先算法进行轮廓遍历,确定最长轮廓为手势轮廓;通过计算轮廓断点的斜率和分析最长轮廓中断点的相对位置,采用局部sobel算子或直线连接断点。实验结果表明,所提方法不仅能够得到优于其他方法的主观视觉效果,而且客观指标也明显优于现有轮廓提取方法,其中轮廓完整性可达98%、轮廓纯净性接近0。Continuous contour extraction plays a significant role in pattern recognition and reconstruction of 3D image. However, the existing contour extraction methods cannot obtain the continuous contour, so it will influence the subsequent image quality seriously. The continuous contour extraction algorithm of gesture is proposed in this paper. Firstly, the Gaussian mixed model is utilized to locate the gesture region in the video image. Then Sobel operator is employed to extract the rough contour of gesture in the selected region. Next, the depth-first algorithm is used to realize contour traversal, and ascertain the longest contour as the gesture contour. Finally, by calculating the gradient and analyzing the relative position of the breaking point in gesture contour, the local Sobel operator or line connection will be adopted to connect the breaking point. The experimental results have shown that the proposed algorithm can get better subjective visual effect and objective performance standard than the traditional methods. Contour integrity can reach to 98% , while contour purity is close to 0.

关 键 词：连续轮廓提取 最长轮廓 手势识别 高斯混合模型 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]