使用臉部訊息輔助自動化膚色偵測

Abstract

膚色偵測被用於許多的影像處理技術，像是臉部偵測或是姿勢偵測等等。由於膚色會隨著不同光源環境以及不同種族而有所變化，造成使用膚色偵測準確性會降低，因此有許多利用前處理像是曝光修正、白平衡或是膚色再偵測等等的技術用來提高偵測效果，然而自動化的調整結果不一定理想，原因在於推測正確顏色需要夠多的資訊，而自動化調整無法得到哪些是正確的訊息。 本篇論文為了避免受到光源影響採用非色彩部份的資訊來輔助偵測。主要方法是用顏色作為膚色偵測的基礎，亮度作為去除非膚色的部份的方法，加上使用臉部偵測用來尋找臉部位置。將辨認出真正人臉的顏色取出，利用臉部的顏色來推斷出整體的膚色，進而提高偵測率。實驗結果將針對各個種族分析使用亮度的優缺點以及有臉部偵測的差別性和色彩修正的差別性。

Skin detection was used in many image processing technologies, such as face detection or gesture recognition and so on. Due to the skin color will be changed by in different light environments and in different races, the skin detection rate will be decreased, so there are many technologies to increase the efficient, such as preprocessing which is like white balance, color correction and etc. , or adaptive skin detection. However, automatic adjusted results may not be correct. The reason is that the correct color needs enough information. The adjustment can not be automated which is the right information. In this paper, in order to avoid the effects of light, uses achromatic feature to increase the skin detection rate. The main method is used color to be the basis, luminance as a way to remove noise, coupled with using face detection to find the location of face. Identifying the true human face color, and uses the face color information to detect skin, it will improve skin detection rates. The experimental results will be used for analysis of all races as well as the advantages and disadvantages of brightness and to detect the difference between the face and color differences of the amendment.