Face Detection In Color Images

Automatic Face Recognition is an advanced subject in the biologic recognition at present. One of its key technologies is face detection.With the development of pattern recognition and computer vision, how to use color information effectively to improve the quality of face detection has become keystone of research. The main research on face detection in colorized images is as follows:Firstly, after several frequently used color segmentation methods were described in the paper, considering the disadvantages of the ellipse segmentation in YCbCr color space and the simple threshold segmentation in HSV color space, an improved algorithm using a combination of the two above methods was proposed, which took into account both the efficiency and the accuracy.Secondly, based on the segmentation results of the skin color, combined with mathematical morphology processing and the connected region analysis, area associated characteristics etc.,the face image screening of candidates obtained better detection result. Finally, experimental outcome of the algorithm was given definitely and analyzed deeply.Thirdly, after analysis and comparison of several commonly used methods of the illumination compensation, in view of the deficiency of traditional methods which have too much color interference in the background and easier to make images be highlighted,a new illumination compensation method was put forward based on the theory of the linear transformation and the index variation, which effectively restrained the color offset and enhanced the segmentation accuracy.Further, eye location was obtained a deep research based on the face location. A new efficient eye-orientation algorithm was proposed, according to the color and brightness information of eyes in the image and combined with the geometrical characteristics of eyes, which had a good result for the eye location.At last, due to the disadvantage of the face detection method based on the skin color which is easier to be affected by brightness or the analogy region of face, while The Support Vector Machines(SVM) has a low detection rate, a new detection algorithm was proposed based on both the skin color segmentation and SVM. During the inspection process,the sample feature was extracted by The Principal Component Analysis (PCA) to reduce the effect of brightness or the analogy region of face.What's more, the Pyramid strategy was used to search the candidate area from the skin color segmentation, accelerating the effectiveness evidently.