Face recognition system is widely used in today's society.Whether it is the identification of suspect photos in security system or the facial unlocking function of smart phones,it brings great security and convenience to people's lives.Face detection function plays an important role in this system.In order to better identify and analyze face objects in images,it is necessary to separate them from images,on which basis,further analysis and processing can be carried out on the target.Face detection is an important process to distinguish "face" from "non-face" images,this process is to find and locate the position and size of the face in the image containing the face,detect the face features of the face and ignore the background or other irrelevant interference.In this thesis,the present face detection algorithms are simply summarized and analyzed,and the advantages and disadvantages of various face detection algorithms are compared,a face detection method based on skin segmentation and a face detection method based on Adaboost algorithm are deeply studied and improved.The advantages and disadvantages of the two algorithms are comprehensively compared,and an improved algorithm combining the two is proposed to improve the system performance.The main tasks completed in this thesis include:Firstly,this thesis gives a concise description of the background and significance,then by consulting the relevant literature,discusses the development status of this research at home and abroad.Finally,we summarize and discuss the face detection methods that people often use at present.(2)Secondly,a face detection method based on skin segmentation is studied.In the use of colour for face region selection stage,this thesis introduces the several kinds of color space and color model commonly used,such as RGB、YIQ、YUV、YCbCr,and then analyses the reason of selecting YCbCr color space and Gaussian model,design the skin segmentation process and screening for skin areas,and to improve the segmentation algorithm.(3)Then,the face detection method based on Adaboost algorithm is further studied.We introduce the basic principle of the algorithm and related concepts,including Haar-like rectangle features,integral figure,etc.And then analyses the weak classifier,the strong classifier and the composition of cascade classifier and the training process.Finally,the Adaboost algorithm was improved to solve the problems in the training process.(4)Finally,according to the advantages and disadvantages of the two algorithms,In this thesis,the traditional Adaboost algorithm is improved,a new haar-like rectangle feature is introduced,and a method combining the two is proposed to improve the system performance.The advantages of this method are as follows: First,the skin color is used to detect the face,which excludes most of the non-face areas in the image.In the detection,the influence of light on the single skin color feature is reduced,and the detection rate of the system is guaranteed.Then,the Adaboost algorithm was used to screen and classify the areas screened in the first stage,and finally determine the face area. |