The Research Of A Convergent Random Forests Algorithm For Faces Detection

As one of the most important components of image processing, face detection has become a focus of correlational researches. There are various methods for detection, making the research appears multi-level and multi-angle. How to use different althorithms effectively to solve detection problems with different standards has become the primary coverage of the research on face detection. We described the essence of detection from information theory and methodology, compared different althorithms in detail and proposed an improved random forests althorithm for face detecting. The main research work shown as follows:1.Introducing different ways for face detection in the principle level, further more did we make comparisons among these ways.2.Describing the random forests in detail and analysising its fundamental theory in detail. Discovered the deep connection between random forests and Adaboost, providing theoretical basis for the improvement of random forests and experiment.3.Redescribe random forests using additive-forward model is used to make random forests grow layer by layer. By the mean of steepest gradient to optimize the process can we get a new random forests called ? ?? RFs. The new model could solve the problem of reaching global optimum solution to some degree, improving the detection speed and efficiency, on the promise of succeeding the advantage of the original model. At last we made experiments that verified the superiority of the new model.4.Characterizing faces using Haar feature. Firstly, generate candidate set through Adaboost, then train sample set using ? ?? RFs. In this way can we enhance the allocability of the althorithm, optimizing our experiment indirectly.