Research On Human Faces Detection And Enhancement Algorithm

Face detection has wide uses in many fileds. When we take a photograph by the digital camera, human faces are detected with difficulty because of the insufficient light or dark circumstances. The purpose of this study is to use image processing method to detect the face in order to improve the qualification of pictures. This problem can be divided into two steps: face detection and face enhancement. To identify the human face effectively, AdaBoost algorithm which is based upon OpenCV is applied to train the classifier. Before enhancing human faces, face region must be extracted exactly according to the skin color information by skin color method. Imadjust algorithm is simple and important in image enhancement based on gray-scale's linear transformation. The Retinex algorithm is a human perception-based image processing which mainly compensates the images for the serious impact of light.In this paper the theoretical basis of AdaBoost is illustrated in detail, integral image, learning algorithm based on AdaBoost and combining classifiers are introduced respectively. We focus on how to train the classifier. The linear transformation equation by Imadjust algorithm is given. The Retinex algorithm can be divided into SSR, MSR and MSRCR. This text improves the Imadjust algorithm and MSR algorithm and divides pictures into R, G, B three channels to realize the image enhancement separately.The whole program is implemented using VC++6.0 and OpenCV (Open Source Computer Vision) executes on the Window Xp. Experiments are conducted by using 297 faces in dark background,253 faces are found, the accuracy of detecting algorithm is 85.2%. Experimental results prove the validity of the algorithm, and also prove the algorithm has fast detection speed and short development period. It is feasible on executing. The Imadjust algorithm can improve the illumination of images well. The MSR algorithm shows the resulting images have better color fidelity, using this method to implement image processing can get very good effect. At last we provide two groups of pictures to show the effectiveness of the Adaboost algorithm, Imadjust algorithm and the MSR algorithm that we improved. The approach algorithms have been proved to satisfy real-time requirement. In the future, algorithms will be applied in embedded system.