基于颜色模型和纹理特征的肤色检测系统研究

【作者】 钟文涛；

【导师】 杨国亮； 周伟扬；

【作者基本信息】 江西理工大学 ， 控制工程， 2012， 硕士

【摘要】 随着肤色检测技术的发展,皮肤自动检测在图像内容识别和过滤,视频压缩以及基于内容的检索等方面发挥着十分重要的作用。而基于颜色或纹理的皮肤检测技术具有简单、快速、直观,不受物体形态改变及视点变化等影响的优势,因此受到学者们的普遍重视,并且具有重要的理论价值和应用价值。本文就是在二者检测技术的研究基础上,结合二者的优点构建了基于颜色和纹理的二级检测系统,使图像针对肤色的检测精度更高。本文具体是针对颜色空间模型和纹理特性对图像中肤色的检测展开具体的研究。主要做了如下一些工作：1、根据颜色空间的肤色检测技术,首先对各种颜色空间,比如RGB、HSV、HIS、YcbCr等进行了介绍,比较各种颜色空间对肤色的检测效果。2、建立颜色空间的肤色检测模型。基于一个大肤色样本库,统计训练出肤色模型需要的参数值,采用高斯模型对肤色进行检测,同时对图像进行光照补偿方法、均值滤波等处理,构成了一个完善的肤色检测体系。采用Matlab进行实验仿真,在实验过程中对不同种肤色分别进行了仿真。通过实验结果表明,基于改进的YCbCr颜色空间的肤色模型检测效果要好于传统的颜色空间,并且该算法可适用于较广范围的光照条件,可以有效地从复杂背景的彩色图像中检测出肤色区域,但还是不能避免类似肤色区域被误判。3、针对颜色空间的检测效果和存在的缺陷,提出一种使用灰度共生矩阵的皮肤纹理检测模型。在上一阶段对图像中皮肤区域用肤色模型进行初步检测的基础之上,首先以训练图像中的皮肤像素为中心构成8×8分块,然后基于一个大的皮肤纹理样本库,提取该样本库中所有以皮肤像素为中心8×8分块的特征值,从而统计训练出图像纹理模型的参数直,并根据该纹理模型对图像中每一像素点进行判定,最终确定有效的皮肤区域。通过实验表明基于颜色空间和纹理特性的二级图像肤色检测系统,使图像中肤色区域的确定更加精确,可以有效避免类似肤色的区域被误判。更多还原

【Abstract】 With the development of the skin color detection technology, automatic detection of skin in the identification and filtering of image content, video compression and content-based retrieval has played a very important role. And the skin detection technology based on color or texture has lot of advantages such as simple, fast, intuitive and not affected by morphological and viewpoint changes. The skin detection based on color and texture has important theoretical and application value, so it has been paid a lot of attentions by many scholars. In this paper a skin detection system based on color space and texture features was implemented, and the skin area can be detected more accurately using the system.In this paper, the skin detection technology was studied in details according to the color space model and the texture features.The main contributions are as follows:1、Some color models which are often used in skin detection, such as RGB、HSV、HIS and YcbCr were introduced and compared comprehensively.2、A skin detection model based on color space was proposed. First, the parameter values of the Gaussian model used in skin detection was learned from a large picture library, then the area of skin in the images was detected. A complete skin detection system was implemented. In the system, the Gaussian model, was used to detect the skin. The image illumination compensation and mean filtering technologies were used too. Simulation experiments for different color skins were introduced on MATLAB. The experimental results have shown that the effect of the system based on YCbCr space is better than the effect of the system based on the other color space and the algorithm is applicable to a wider range of lighting conditions. The proposed algorithm can detect the skin areas effectively from the color images with complex background.But the algorithm cannot avoid similar color regions being detected wrongly.3、A skin detection model based on the GLCM is proposed to improve the detection results of the color space model. All the work in this stage is based on the results of the work completed in the before stage. Images, were divided first to several image blocks whose central points are the skin pixels, and the size of the blocks is8×8. The parameter values of the skin detecting model was learned from a training skin library. Each pixel was identified using the skin model and the effective skin regions was determined in the images at last.The experimental results have shown that the proposed two-stage skin detection system based on color space and texture features can detect the skin color area more accurate and can effectively prevent similar color regions being detected wrongly.更多还原