Font Size: a A A

Study Of Eyebrow Recognition Base On AdaBoost Detection Method And Sub-Region Matching

Posted on:2012-03-01Degree:MasterType:Thesis
Country:ChinaCandidate:H J LiFull Text:PDF
GTID:2178330338991443Subject:Computer Science and Technology
Abstract/Summary:PDF Full Text Request
Along with the in-depth development of nowadays informational society, biometric technology has become not only a hot field at home and abroad, but also one of the key strategic technologies competed by states. Compared with traditional identification methods such as cards, keys, accounts and passwords, biometric technology is more convenient. At present, the biometric technologies used commonly include fingerprint, face, iris, palm, gait, ears, voice, handwriting and so on. Theoretically, any biological characteristics, including physical characteristics and behavioral characteristics, which are universality, diversity, stability and collectable, can be used for individual identification. However, the eyebrows, which also have such features and perform an important role in the face, are rarely used as an independent biometric for recognizing. Based on the pre-works, this paper does some further research in the eyebrow segmentation and recognition. The main works are described as follows:1) We study location and segmentation methods of pure eyebrow regions. Firstly, we locate the rough position of the eyebrow using the AdaBoost cascade classifier, and the position is set to be the initial seed region of the sub-regional growth method. And then, the boundary points of the eyebrow can be found using the sub-regional growth method. Finally, the pure eyebrow region can be selected by a polygon which is constructed using the convex hull method. This approach can get good results under the high-quality image, but will be affected by the binary image.2) We study a fast method of constructing eyebrow feature string. In the eyebrow recognition method based on the comparison of feature string, minimal spanning tree (MST) is used in the process of constructing eyebrow feature string. When the dataset is large, using the classical Dijkstra algorithm for MST will get a low efficiency. So a fast spanning tree method based on the Locality Sensitive Hashing, called the Hashing Dijkstra Fast-Approximate algorithm, is proposed. This algorithm uses secondary hash index tables to accelerate the searching of neighbor points, so that to construct a tree fast. Though it is an approximate minimal tree, the time of constructing is lower than the classical Dijkstra algorithm. Experiments on the eyebrow recognition with 109 persons show that the method using the fast algorithm got the recognition rate at 81.65%, which is little lower than that of using classical Dijkstra algorithm (about 83%).3) We propose a novel matching-recognizing framework, and then study eyebrow recognition in the framework. The eyebrow recognition method in the matching-recognizing framework consists of two processes: matching and recognizing. In matching process, the fast template matching method can be used to locate the target region by normalized cross correlation as a matching similarity; and in the recognizing process, the final identity is determined via Fourier spectrum distance as a discriminative similarity. A number of experiments were executed on the opened BJUT Eyebrow Database (BJUTED), and this method got a high recognition rate which is up to about 97%. Moreover, only replacing the eyebrow templates with face templates, this method can be used in face recognition directly. Experiments on Color FERET Face Database show that the eyebrow recognition rates are higher than that of face (when 100 subjects, eyebrow recognition rate was 85%, while face recognition rate was 81%; and when 800 subjects, eyebrow recognition rate was 75.66%, while face recognition rate was only 72%). Those not only verify the feasibility of the eyebrow recognition, but also powerfully illustrate that the eyebrow recognition can replace face recognition in some case.
Keywords/Search Tags:Biometrics, Eyebrow Recognition, Matching-Recognizing Framework, Template Matching, Image Segmentation
PDF Full Text Request
Related items