Face Image Retrieval Using Pose Specific Set Sparse Feature Representation

There are largely available consumer photos are available in our life, among those a big percentage of photos are photos with human faces. Thus content based face image retrieval is an emerging technology for many applications. A new method for content based face image retrieval is proposed. Given a query face image, content based face image retrieval find similar faces from the database. There were mainly three steps- preprocessing, feature extraction and face retrieval. Preprocessing included face detection and landmark detection. After preprocessing, extract pose specific uniform LBP features from the face. Novel set sparse feature representation is used to represent extracted pose specific LBP feature as sparse code using a pre-defined set. After 15 attributes from face is used to make face representation more distinguish. Finally set sparse feature representation is used as inverted index in face retrieval to find similarity score of query face image with database face images. It has applications in automatic face annotation, crime investigation etc. Top results related to a query face image with existing method and memory usage were analyzed. Experimental results shows that proposed method have better top results and efficient in memory usage compared to existing method.