Color Image Face Detection Feature Location Technology Research

Face detection is a discipline related to pattern recognition, image processing, physiology and other fields. The initial study of face focused on face recognition, with the expansion of face application, computer technology matures, people's growing demand for practical applications, the human face detection divided from the face recognition as an independent research project receiving more and more attention. At present, domestic and international has emerged a large number of face detection methods, but the existing algorithms can only apply to a certain environment. However, due to faces'own complexity and the external environment and other factors made it difficult to detect face. Therefore, to achieve a high detection rate and good adaptive auto face detection system under an unconstrained environment is an extremely challenging task.This thesis achieves the color segmentation using the color-based face detection algorithm, taking the complex face images of multi-faces and simple face images of single face as the research objects, and achieves the key feature location in the simple face images of single face. The main contents are as follows:(1) Analysis the various color space of the computer vision areas, taking into account skin color only related with the Chrominance and not related with the luminance, we select the brightness and chrominance detachable color space YCbCr, and adjust brightness of the image before using the skin color for face detection. By comparing the brightness adjustment methods such as "Reference White", logarithmic transformation, this paper uses the compensation method based on "Gray World", it can directly achieve fast operation on RGB images.(2) As the YCbCr color space, luminance is not entirely independent of the color information, this paper will convert YCbCr color space to YCb'Cr'color space by nonlinear transformation, and then use the color similarity model and the elliptical model respectively to achieve face-color segmentation for the complex face images of multi-faces and simple face images of single face.(3) Because of the noise involving in the color segmented region will affect further face locate, this paper uses a pixel-based'density'method of filtering by gradually narrowing the scope of neighborhood, and makes use of the mathematical morphology opening operation to achieve noise processing on the segmented face-color region.(4) At last, this thesis achieves location of key feature points to the simple single-face images. This method was based on knowledge of the features localization. Before executing the location, it divides the face region into separate areas according to the distribution rules of human faces, it narrows the search range and can achieve rapid positioning. The method was tested in the test set established by our own, but there still exist false detection, undetected and a certain bias of key feature location owing to the complexity of the face own and the impact of changes in posture and expression and so on. Therefore, how to reduce the false detection rate and false negative rate, and improve face detection accuracy, while achieving key features location of multi-faces images of complex background will be the next focus of our study.