Robust Face Recognition Using Histogram of Oriented Gradients and Auto-Associative Memory Network

Facial recognition technology is based on a person's facial features. It is often the core of applications such as biometrics, access control, and forensics. However, due to environmental factors such as dust accumulation on the camera lens or equipment failure, the accuracy of the system is often questioned; especially when it comes to noise and partial occlusions in the test images. This paper proposes a new approach for automatic face recognition using histogram oriented gradient features of facial images with auto-associative memory in the form of a “Multi-Connection Hopfield Neural Network”. The method described here efficiently stores and retrieves facial patterns in noisy or partially occluded environments. In testing, this proposed approach achieved a recognition rate of 98.5% on the ORL dataset, outperforming these other state-of-the-art methods. Furthermore, this method maintains 95.0% accuracy even with 50% random noise and exhibits superior performance and robustness under partial occlusion and variable lighting conditions.