摘要： 实现了一种用于说话人识别的实时唇线提取方法。该方法首先将原始图像变换到CIELAB 颜色空间，然后对所得图像的L 和a 分量分别进行直方图统计。根据图像特性、L 和a 坐标轴的意义等先验知识，直接快速地计算出用于分割出唇区域和嘴内区域的阈值，达到较准确的唇定位。然后采用新的能量函数指导变形模板法提取出唇线。实验表明，该方法简单快速，能够在一定程度上适应光照变化，以及由于不同人所带来的肤色、唇色不同的影响

关键词: CIELAB；唇线定位；图像分割；变形模板；直方图

Abstract: This paper presents a new method of real time lip contour extraction for multimodal speaker identification. Firstly, it turns the original image into CIELAB color space, then computes histograms of L and a separately. According to the characteristic of the original image and the meaning of L and a coordinates, the paper can work out the threshold for segmenting inner lip area and lip area. Then lip contour is extracted using improved deformable template. This method is simple and fast. It turns out to be robust to illumination and speaker independent as experiments show.

Key words: CIELAB; Lip contour location; Image segmentation; Deformable template; Histogram