Multi-view Face Detection Based On Cascaded Adaboost Classifier

As a kind of important biological feature, human face has broad prospects and application value in the field of computer vision and pattern recognition, which has attracted the interest of many researchers. However, human face is also a complex non-rigid object, which is full of challenges. And the process of multi-view face detection is much more difficult. Although the current method has achieved good results, there are still some limitations. Against this background, the paper has studied the adaboost algorithm, which has gained more concerns in the areas of face detection in recent years, and has proposed a fast multi-view face detection method based on cascaded adaboost classifier. The main contents are as follows:1. In the face detection process, we used the idea of detecting from coarse to fine. First, we used detection method based on objectness to obtain the proposal area of human face. Then, we used the haar-like features and trained face classifier based on adaboost algorithm to detect faces precisely in the proposal face area. By using this method, we get the face region rectangle in an image with less sliding windows and reduced detection time significantly.2. While training a face classifier of a specific view in the training process of multi-view face classifiers based on cascaded adaboost algorithm, we has collected other point of view of face images and background images as negative samples. And we gave more attention on the negative samples of face image so that they can be excluded as soon as possible with a high probability.3. We has proposed multi-level cascade structure for multi-view face detection after we had gained classifiers with better classification, which has accelerated the process of face detection.4. We have done experiments on the CMU profile face test set. And the results show the validity and accuracy of this method...