Gravitational search-based optimal deep neural network for occluded face recognition system in videos

Video surveillance is an effective method to improve public safety and privacy. Video surveillance technology has entered a stage where increasing video cameras are inexpensive but requiring open staff to evaluate the videos is expensive. The extensive researches conducted using machine learning for automatic face recognition doesn't provide accurate results as of human evaluation. To enhance the biometric features of the security applications, automatic face recognition is used. Surveillance records incorporate various challenges for face recognition and face detection. For instance, facial recognition systems can be affected by the size of a face image, occlusion, posture, lighting conditions, and establishment, while recognition accuracy may be affected as a result of low objectives, occlusion, posture, light, and dimness. To conquer these obstacles, an effective face detection and recognition framework is proposed with optimal feature extraction methods. At first, the keyframes with face pictures are extracted using a strategy known as keyframe extraction using wavelet information. After the extraction of keyframes, the multi-angle movement feature, SURF feature, holo-entropy, and appearance features are used for feature extraction. Finally, the recognition can be done using optimal deep neural network based on the gravitational search algorithm. Therefore, the proposed method's performance is evaluated using various benchmark video dataset. The efficiency of the proposed approach is evaluated by comparing it using sensitivity, specificity, accuracy, keyframe extraction time, etc.