基于自训练与snakes搜索的活动形状模型  

Self-training and snakes searching based on active shape models

作　　者：朱才志[1] 周晓[2] 吴秀清[1] 张垒[1] 

机构地区：[1]中国科学技术大学电子工程与信息科学系,安徽合肥230027 [2]合肥工业大学计算机科学与技术系,安徽合肥230009

出　　处：《中国科学技术大学学报》2007年第9期1106-1112,共7页JUSTC

基　　金：中国高技术研究发展(863)计划项目(2004AA783052);十五总装预研项目(42201020501)资助

摘　　要：活动形状模型(ASM)算法在有模型监督的轮廓提取中应用广泛,其不足主要体现在两个方面:(1)对大量样本标记点的标注费时、费力,且易产生误差;(2)算法轮廓进化中的盲搜索对图像中的噪声敏感,容易偏离全局最优.对前者,当目标物体的待成像轮廓在同一平面时,可利用透视投影变换模拟产生物体轮廓在3D空间成像的训练样本,从而完全避免手工标注引入的人为误差;对后者,在ASM的轮廓进化过程中,结合经典的snakes活动轮廓模型算法(ACM),可提高算法收敛的鲁棒性.上述改进的ASM算法已用于视频相册系统中,试验结果证明了算法的有效性.Active shape model （ASM） approach is often used in extracting the contours of objects when their contour models are known in advance. The main shortcomings in ASM lie in. （1） Manual labeling of landmarks in a large training set is a rather hard work and is prone to man-made error. （2） Blind searching in contour evolution is sensitive to noise of image data, and therefore tends to deviate from global optimization. For the first defect, if the contours of the target object which should be captured are planar, the perspective projection transform can be used to simulate training samples of the contour of the object captured in 3D-space, thus manually labeling process as well as man-made error is completely avoided in our approach. In order to overcome the latter shortcoming, an active contour model （ACM） based contour evolution process in ASM was proposed, which can efficiently enhance the robustness of the approach. This improved ASM approach has been applied in the video booklet system, and the experimentation results proved its validity.

关 键 词：活动形状模型 自训练 活动轮廓模型 轮廓提取 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]