Face Detection Based On Image Fusion Algorithm Of Wavelet Transform

With the development of society and technology, the request of fast, effective autotmatic face recognition becomes urgent. Biological characteristic is part of human's intrinsic attribute which has a good self-stability and individual difference, has been enormously valued and developed in the domain of scientific research. As a key step of face recognition, face detection has been treated as an independent topic and becomes a hot research spot.The technology of face detection has been developed for many years and accumulated a lot of research results, achieved the very high level. But when the background is complex, or the face image is fuzzy and distorted, the effect of face detection is not ideal. Image fusion method can take useful information from several face images and merge to one, which has good clarity and contrast. Apply image fusion method to face detection can decrease errors of face detection to half-baked face information.Prime task and achievement of this paper are as follows:1. Take image fusion method of video images basd on DirectShow to face detection. DirectShow can make application programmers get rid of work of complex data transmission, hardware difference, synchronism and so on. Using directshow to develop real-time preview and face detection of video becomes easy.2. Pretreatment to video images which would be fusd. Pictures taken from different angles often contain complementary information and certain degree of distortion. Correcting pixels of distorted image to resume the color and space relationship it used to be. 3. Propose a method of face detection basd on local entropy of image fusion algorithm via wavelet transform. Apply wavelet decomposition to video image. Because the low frequency is close to the original image, the pixels correlation is not very strong. Take the simple method of average , the wavelet coefficients can be obtained easily. The brim,detail and so on are contained in high frequency, is basd on the coefficient's neighborhood entropy. The fusd image was reconstructed by performing inverse wavelet transform to fusd coefficient matrix.4. Improve the training speed of weak classifer of AdaBoost. Place the feature value of all samples from small to large, to form an ordered list. Take each item of the list as a threshold and calculate corresponding error. Take the threshold which have the least error as the threshold of the weak classifier.Pretreatment to images which have complementary information. and then merge to one fusd image for face detection, and got a good detection result, increased the detection rate of face detection algorithm in distorted faces. Face detection is a ever-developing topic, how to detect face exactly and fast worth our further research.