Ms.Reeniya Bopaiah A, Mrs.Rajeshwari J
Department of Information science and Engineering
Dayananda Sagar college of Engineering, Bangalore, India-560078 Abstract
Face recognition is an important security application for automatically identifying and verifying a person from image or video source. Face recognition under partial occlusion is a challenge in the field of image analysis and computer vision. Although face recognition has been encouraging, the task turned out to be difficult when illumination, …show more content…
It can easily identify individuals from massive crowed.
The system is user friendly and does not require the contact of a person for authentication. The face recognition methods are classified into the following category:
Holistic method
Structural method
Hybrid method
In holistic method the entire face is considered as input that is the entire face features are considered for recognition. One of the best examples of this type of method is eigenfaces and principal component analysis. In structural method selected local features such as eyes, nose and mouth are extracted for their location and local statistics are fed into the classifier. The structural method is further classified into generic, template based and structural matching method. Hybrid methods are the combination of holistic and structural methods and are generally used for 3D face …show more content…
The matching is then performed based on relational vector space which consists of feature correspondence and structure similarity. They used AR face and Hong-IK face database and obtained an accuracy of about 97.73%. Jongsun Kim et.al[2] proposed a part-based local representation called locally salient ICA for face recognition under partial occlusion. Here part-based local basis of image were created using localization method such as LNMF and local feature analysis(LFA). They used FERET, AR face and AT&T database and obtained an accuracy of about 75-80%. Xiaoyang Tan et.al[3] proposed a method where they used self-organizing map(SOM) to represent each individual face image, then they used two strategies for learning the SOM topological map where a single SOM map was trained for all samples and then a separate SOM map was used for each class. They then used soft K-nearest neighbor classifier for recognition from SOM topological map. They used AR face database and obtained an accuracy of about