基于改进的level set嘴唇轮廓定位方法

根据嘴唇的几何分布特性,提出一种基于多方向的水平集方法（Multi-Level set）来进行嘴唇轮廓的定位。Multi-Level set方法通过对嘴唇图像多个方向的滤波得到新的边缘检测函数来增强嘴唇轮廓的梯度信息,然后利用能量函数最小化来使初始曲线向嘴唇轮廓靠近,达到说话人嘴唇轮廓的精确定位。实验证明用Multi-Level set方法定位嘴唇轮廓的准确率比level set提高了7.32%。

Improved level set-based approach for lip contour detection

To detect lip contour, a multi-direction level set approach (Multi-Level set) was proposed for detection according to the lip geometric distribution. Multi-Level set approach could strengthen the gradient information of the lip contour by using mutil-direction filters on images. And then the initial contour of lips was made to evolve to speak to people around the lip actual contour by minimuming energy function. Experiments on mouth images show that the improved accuracy of the shape contour detection using Multi-Level set approach is nearly 7.32% over level set approach.