AdaBoost Algorithm For Face Detection And Hardware Implementation

At present, the human face detection, which is to detect a prepared image todetermine whether it contains face, the size, the location and other related informationby the means of computers, is the highlighted subject in the field of pattern recognition.Now that it has been applied widely into practice such as security, the companyattendance, video monitoring and digital image processing etc. a great number ofpersonnel undertake the research at home and abroad. But there exists difficulty in thedetection due to the following factors. A human face is the3D non rigid target with thestructure constant such as nose, eye, ear etc; the appearance variability like expression,looks, illumination and so on; the pattern variability like image resolution, images ofgeometric variation etc and complexity and diversity of the facial features such asinstance, color characteristics, elliptical contour, symmetric and so on.The main contents of this article are as follows:To begin with, it detailedly analyzes the basic principle of several traditionalAdaBoost algorithms as well as its application in the human face detection and safelyconcludes its weakness. Secondly, it puts forward the corresponding improvementmethods. From the above explanations, we can naturally come up with the fast facedetection algorithm as a result of combining the stratification of PCA features withAdaBoost algorithm. Finally, it designs a hardware diagram to illustrate the fast facedetection algorithm.The main innovations of the paper are as follows:1, To replace the traditional Haar feature with Walsh feature for face detection and useenhanced Cascade classifier instead of traditional Cascade one. Walsh feature havingmutual orthogonaylity of feature vector, there will never produce redundancy whencalculating the human face, which contributes to save time and the computer’s storagespace, thereby, to improve the detection speed. Experiments show that the result in thisway is identical to that by Haar feature. Compared with the traditional Cascadeclassifier, the enhanced one has the advantages of keeping the training sample’sautonomy as well as the sample classification information from the front layer merelythrough adding an evaluation standard between the front layer classifier and itsadjacent back layer classifier. Although this will increase the computational difficultyof face detection, detection precision is improved to some extent 2, To combine PCA with AdaBoost algorithm. In the former phase of face detectionWalsh classifier is employed instead of traditional Haar, and enhanced cascadeclassifier is a substitute for traditional cascade at the same time. Since Walsh is unableto detect the image exactly owing to its inherent shortcomings, for this reason, PCAfeatures and AdaBoost are combined to examine the pictures of the human face in thelater phase. Experiments show that this hierarchical face detection algorithm canachieve face detection rapidly and accurately.3, To design the hardware. Because of the calculation complexity of this hierarchicalface detection method, there is specially system requirement and the traditionalARM11can not meet the need. As a result, embedded DAP processor and ARM9DM6446in TI as the main chip are applied to construct a system for face detection.With the help of all aspects, the practice proves that this system is stable and cancomplete the human face detection fast.