一种基于Kinect深度图的人像自动抠像算法

An Automatic Matting Algorithms of Human Figure Based on Kinect Depth Image

针对传统抠像算法需要人工交互,无法实现自动抠像,且边缘区域前、背景颜色相近时抠像效果不佳等问题,研究一种基于Kinect深度图的人像自动抠像算法。利用Kinect人物识别的优势,研究根据采集的彩色图像对于人形区域深度图像进行预处理,生成三元图;对人形区域中孔缝进行处理;采用改进的鲁棒抠像算法实现无需人工输入的自动抠像。实验结果表明,抠像算法可实现自然背景中人像的自动抠像,对于前景和背景边缘颜色相近的原始图像,可得到良好的抠像效果。

Traditional matting algorithms requires manual interaction in order to separate the foreground and the background. And when color of pixels in the edge from both foreground and background are similar,poor matting effects are obtained. According to these problems,an automatic matting algorithm of human figure based on Kinect depth image is proposed in this paper. Kinect has an advantage of recognizing human figure. Taking both the color image and the human figure depth image as inputs which are shot by Kinect at the same time,the depth image is preprocessed by joint bilateral filter. Tri-map is produced automatically,and tiny holes and gaps are considered. An improved robust matting algorithm without manual input is conducted. The experimental results show that human figure can be matted automatically,and a good effect can be obtained even if pixels at the edge have same colors.