Research on Face Recognition Algorithms in Social Software Based on Deep Learning

With the rapid development of mobile internet technology and the widespread adoption of mobile smart devices, mobile social software has garnered significant attention. Traditional face recognition technologies struggle to achieve satisfactory recognition results under large data volumes and unrestricted conditions. The integration of deep learning technology has improved face recognition performance in such scenarios. This paper addresses the challenges posed by factors such as lighting, angles, and expressions on face recognition applications by proposing an improved Deep Belief Network (DBN) face recognition algorithm. The Local Binary Patterns (LBP) features of images are used as input for training the neural network, enabling the DBN model to exhibit robustness against variations in lighting, angles, and grayscale, while retaining the characteristic of bottom-up self-learning of data features inherent in DBN. Additionally, to tackle the issue of overfitting caused by insufficient datasets during the DBN training process, this paper introduces a Dropout technique to enhance the DBN model, resulting in the DDBN (Dropout-DBN) algorithm. This effectively prevents overfitting, improves the generalization ability of the DBN model, and experimental results from the ORL and Yale databases demonstrate that the DDBN face recognition algorithm achieves high recognition rates under unrestricted conditions, with lower computational demands, making it easier to obtain optimal solutions. Furthermore, even with insufficient data, it maintains good recognition performance.