Fusion of LBP and Median LBP for Dominant Region Based Multimodal Recognition Using Imperfect Face and Gait Cues

The paper provides a novel approach to recognize imperfect face and gait cues by using only the dominant region of such cues which contain more information than that possessed by the other regions. To enhance features in such dominant region, a fusion of Local Binary Pattern (LBP) and Median Local Binary Pattern (Median LBP) procedure is proposed. Initially, the given imperfect face and gait probe images are divided into six overlapped half regions. After partition, the dominant overlapped half regions of face and gait are selected by using content based image retrieval process. Subsequently, the features of dominant overlapped imperfect face and gait regions are enhanced by fusing the feature vectors obtained by using fusion of LBP and Median LBP methods. Next the Eigen feature vectors followed by Fisher’s vectors are constructed using Principal Component Analysis (PCA) followed by Linear Discriminant Analysis (LDA) dimensionality reduction algorithms. The decisions from face and gait biometric systems are obtained separately by using Euclidean distance measure. Finally, the decisions of face and gait classifiers are fused at decision level for recognition using AND/OR rule. The method is verified on freely available ORL face and CASIA B gait datasets.