The Research On AdaBoost Face Detection Algorithm Based Improved Skin Color Model

Face detection is the first step of advanced information processing algorithms such as face recognition, expression recognition, which the specific work content is through certain test order and test algorithm search target image in the video(or target) whether they contain face feature or not, if there is contain face feature it returne face location and size information. The effect of the follow-up work such as the correctness and detecting speed are directly affected by the test results. Therefore as the field of computer vision and artificial intelligence development, manifest the important position of face detection, and face detection also gradually become a hot topic in study.Existing face detection algorithm are numerous, but the disadvantages of them are more obvious, mainly reflected in the detection rate isn’t high enough, the real-time perform is so poor of algorithm, and easy to affected by the external environment and the change of main detect body and so on. In view of these disadvantages, a new face detection algorithm on the basis of the several classical algorithms is proposed in this paper, which combining skin color model and Ada Boost algorithm. Simple skin color model is used as the weak classifiers of Ada Boost in this method, and according the rules of Ada Boost algorithm to training and testing these weak classifiers, then achieve several strong classifiers in order to realize the accurate location of face. The main work of this thesis can be summarized as follows:(1)Research of face detection algorithm based on skin color model. The skin color clustering effect in different color space and the fitting effect of different skin color model are analysed in this chapter. The testing results of face detection method based on skin model is showed and analyzed, at the same time, the work of image preprocessing method are introduced, such as optical compensation method, normalization of size, etc. Finally, the test results are improved and processed by mathmathematical.(2)Research the face detection method based on Ada Boost algorithm. The rules and operation process of Ada Boost are analysed, include the structuring method and training rules of the weak classifier, the structuring method and training rules of the cascade classifier, as well as the detection rules of Ada Boost algorithm. Finally the advantages and disadvantages of this algorithm are summarised.(3)An improved Ada Boost face detection algorithm is proposed. Skin model is used directly as the weak classifiers of Ada Boost, so the detection precision of the weak classifier is improved effectively, while the Ada Boost algorithm off-line training time is reduced and the detection rate of the algorithm is ensured. In addition, the weighted voting is used to decide the results to avoid the weight distribution imbalance called degradation. Experimental analysis shows that the detection rate of this algorithm is effectively improved while the false detection rate is reduced. Also the training time is shorter at the same time. And the conditions of detecting object constrint is very low, and the robustness of algorithm is good.(4)For complex background(similar color region, extreme light environment, skin region out of face) interferencing with face detection work, a new method combined visual significant mechanism with improved Ada Boost face detection algorithm is proposed. The skin color is calculated as the aiming figure through this method and gets the saliency histogram, the target location information in the space is ignored, and the significance is proposed as new parameter to strengthen the constraints of the background interference item. By retaining the color of the color histogram distance target recently to exclude obvious regional areas and different colors although similar to skin color but also does not belong to the area of face, the interference of similar color region and extreme light environment in the complex background are reduced. The method is used to test samples and the test result is obviously improved.