肤色检测结合Grab Cut的服装图像前景提取

Foreground Extraction of Garment Images with Skin Color Detection and Grab Cut

为解决全自动化Grab Cut算法应用于服装图像的前景提取时,无法去除模特肤色的干扰以及对一些复杂背景图像存在欠分割的问题,提出了新的前景提取方法:为有效去除模特肤色区域,提出了一种基于双边滤波去噪的肤色检测改进方法;为有效地去除欠分割区域,保留服装前景区域,提出了一种轮廓检测算法;将改进的肤色检测方法、轮廓检测算法与全自动化Grab Cut算法相结合,进行服装图像的前景提取.实验结果表明:改进的肤色检测方法相比于现有的肤色检测算法,肤色检测的准确性明显提高;改进的前景提取算法的服装前景提取效果显著优于全自动化Grab Cut算法.

In order to solve the problem of the interference of the model＇s skin color and the under-segmentation of some complex images when the automatic Grab Cut algorithm is applied to foreground extraction of garment images,a new algorithm of foreground extraction is proposed. In order to remove the skin area of the model,an improved skin detection algorithm based on bilateral filtering for image denoising is proposed. In order to remove the undersegmentation regions and preserve the foreground regions of a garment,a contour detection algorithm is proposed.The improved skin color detection method and the contour detection algorithm are combined with the automatic Grab Cut algorithm to extract the foreground of garment images. The experimental results show that,compared with the other existing methods,the improved skin color detection method increases the accuracy of skin detection significantly and that the proposed foreground extraction algorithm is better than the automatic Grab Cut algorithm in garment foreground extraction.