Integrating Global and Local Features for Efficient Face Identification Using Deep CNN Classifier
Khushboo Jha, Sumit Kumar Srivastava, Aruna Jain · 2023
This work analyses the effectiveness of integrating global (complete face) and local (eyes, nose, mouth part) features for face identification using deep CNN classifier for different percentage of training dataset and different number of users. In this work, we have used HOG for face detection followed by DHE technique for image enhancement as low-illumination image contain important distinct facial features. The role of global and local features in perceiving a face is distinct and substantial. Thus, to examine the efficacy we have extracted and integrated the global feature (DCT) with local feature (ULBP) as the proposed facial feature. For dimensionality reduction of DCT-ULBP feature, 2DPCA is used. The reduced feature vectors are fed to deep CNN classifier for face identification task. The experimental result on ORL face database, shows that the proposed features are discriminative and complement with each other. The proposed DCT-ULBP feature with deep CNN helps us to achieve an accuracy of 94.58% and outperforms the contemporary DCT and ULBP feature extraction techniques.