AI Summary & Key Takeaways
AI InsightExperimental results show that the proposed multimodal system performs better than the unimodal system and other multimodal systems that use different data fusion rules (e.g.
📌 3 Key Takeaways
Abstract
In this paper, we propose a multimodal biometric face recognition technique which is mainly based on the 2D Discrete Wavelet Transform (DWT) and Data Fusion (DF) and utilizes data fusion techniques at the score level of the system algorithm. The technique employs three discrete unimodal feature extraction and classification methods. The first two feature vectors are generated from raw images by using Principal Component Analysis (PCA) and Local Binary Pattern (LBP) methods. During the generation of the third feature vector, images are initially transformed into the DWT domain. In result, approximation, vertical, horizontal and diagonal detail matrices are combined to form a Joint Feature Vector (JFV). K-Nearest Neighbor (KNN) classifier algorithm is separately applied to the three generated feature vectors to compute different score values for the same individual. These raw score values are fused together using a newly proposed data fusion technique based on Manhattan Distance Penalty Weighting (MDPW). The proposed MDPW penalizes an individual for scoring low points and further pushes it away from the potentially winning class before data fusion is conducted. The proposed approach was implemented on ORL and YALE public face databases. The results of the proposed approach are evaluated using the recognition rates and receiver operating characteristics of the biometric classification systems. Experimental results show that the proposed multimodal system performs better than the unimodal system and other multimodal systems that use different data fusion rules (e.g. Sum Rule or Product Rule). In ORL database, the recognition rate of up to 97% can be achieved using the proposed technique.