Research On Face Detection Based On Mathematical Morphology And AdaBoost Algorithm

Face Detection is to determine whether there is someone in a given image, if yes, then figure out the position and scope of the people's face. Face Detection is a difficult task because it is hard to extract the face feature. The face feature will change if the gesture of the person, the light, the imaging condition changes. All this harder the extraction of the face feature. Therefore, all face detection algorithm usually have some problems, like too much calculation, low speed, poor robustness and so on.The main purpose of this paper is to detect the human face with a rapid and accurate method in a given image. This paper uses a composite strategy, including two parts. The first part is to find out the candidate human face,and the second part is to verify the candidate faces. It uses mathematical morphology algorithm for grayscale image in the first part and classifier basing at AdaBoost in the second part.In the process of searching the candidate human face,the gist is that the gray value of the eyes is lower than other parts in a image. Morphological closing operation can filter gray bottom, removing the eyes,then extract out the candidate eyes by subtracting with the original image. Known the candidate eyes the candidate faces can be extracted. In the faces verification part, the classifier based on AdaBoost is used. Experience from previous studies shows that, although the use of the best classifier can have better recognition performance than any other classifier, the errors different classifiers (including best classifier) generating not always overlap, or errors generated by the classifier with the worst performance not always includes that generated by the best one. It shows that the classifiers that is not the best may provide some important complement information to improve the performance of the best classifier.The paper details the two parts of the method, and an experiment is performed with it. The result of the experiment shows that the attempt is successful, fast detection speed and high detection rate is obtained.