Face Detection Technique Using Convolutional Neural Networks (CNNs) and MobileNet

Abstract

Facial detection has become a significant advantage in the field of image forensic and computer vision. These provide help in detecting an image that are doctored in order to hide or conceal information/message. Researchers used extensive deep learning applications particularly those algorithms that highlight potential for detecting facial recognition to address issues such as high computation, long operation time and low accuracy in traditional and lightweight face recognition algorithms, we introduce a face recognition method that incorporates a lightweight neural network and multi-hash recognition degree weighting. Firstly, the front-end feature extraction in the SSD detection network utilizes the improved MobileNet model instead of the VGG model. The pruned SSD model is employed as the detection network, reducing calculation time and preventing overfitting. Secondly, face matching involves weighing the calculated mean hash similarity and perceived hash similarity using a specific weight. The proposed improved method is trained and evaluated on WiderFace, LFW and FDDB datasets. Experimental results demonstrate that the method enhances running speed by ap proximately 29% on LFW datasets while achieving a high face recognition rate and maintaining detection performance. This research introduces a novel and high-performance approach to the field of face recognition.