改进的先验形状模型构造及其人脸图像分割

The Construction of Improved Shape Prior Model and Its Application in Face Recognition

在处理分割被遮挡、背景与目标灰度相似或数据丢失的目标时,需在CV(Chan and Vese)模型的基础上引入先验形状信息。传统的先验形状项只具有旋转、缩放和平移不变性,针对表情丰富、易产生遮挡的人脸图像,分割结果很不理想。结合形状统计的水平集图像分割做了如下两点工作:(1)在CV模型基础上加入局部剪切和X、Y方向拉伸不变两种特性上,建立了新的数学分割模型;(2)构造新的先验形状能量项,对全局变化和局部变化的人脸图像都能进行平滑快速的演化。实验结果表明本文方法对复杂背景下姿态变化较大的人脸图像,具有较好的分割效果。

In dealing with obscured background segmentation and target gray similarity or loss of data, in CV （ Chan and Vese） model on the basis of introducing the prior shape information. Traditional prior shape item only with rotation, scaling and translation invariance, segmentation result is not very ideal in face image that is occluded or of rich expression. Combining with level set image segmentation based on the shape statistics, the following two points are done ： （ 1 ） A new math segment model is constructed ,which joines local invariant to X ,Y direction tensile and shear properties based on CV model basis ; （2） A new construction method of shape energy term that makes evolving surface stable is presented model can achieve good , which considers the global and local change. Experimental results demonstrate that our segmentation towards face images with great pose variation in cluttered background