A Research On Improved PHOD Method Based On Multi-Layered Voting

With the development of social technology and economy, the expression of information is more and more humanization, that is, the expression of information is more and more convenience. Because of the character of easy receiving, images become one of the most important way to cast the multi-media information. There are large semantic informations which can be included in images, but we are not interesting in all of these information. That means, in the area of feature extraction, content-based image recognition and retrieval need to draw large scale of features, but if we can capture the valuable part of the image, the work will be distribute to the following research. So, the image detection become the primary research area. The acute detection of object in the image, can not only eliminate disturb information, but also capture the mainbody part accurately in the further recognition and retrieval work.Face detection, is a face location technology particular using analysis and comparison method, witch will be contributed to the application of face recognition and Authentication. The technology is also a hot area nowadays. Because the technology is relatively mature, there are large amount articles on face detection exist. So we choose face detection and recognition as a main entrance, research the art-of-state algorithms.The Support vector machine(SVM)classification is a essential section in object detection. But during the test, we found it has problems, to resolve the problem, we propose Multi-Layered Voting. This voting can improve the classification result by only using the most effective levels' features. Because in the image edge, there exit larges of structure information, so using the edge as feature can delete much disturb information, also, it can decrease the amount of digit data. At the same time, histogram of gradient orientation can capture the contour and shape characteristics well, so based on the edge image, we use PHOG as the feature extraction operator.These years, Space Pyramid technology has become the core algorithm in image processing. It can effectively retain the spatial information of images, express image semantic more accurately. Therefore in this article, through the space pyramid division, we add spacial information to PHOG. In order to improve the classification result, this article using the Layered Voting to concentrate our attention on the levels has best classification result, and only use these layers' feature to calculate. This article take detection and recognition test on ORL separately,101 object category and UIUC car database, the results show that, no matter the research area is detection or recognition, our improved method can have a better result.