Enhancing Face Detection in Low-Light Conditions: An Analysis of Noise Types and Denoising Techniques

The clarity of a photograph is diminished by many noise forms, including Gaussian noise, Poisson noise, and salt- and-pepper noise, which makes face recognition challenging in low-light conditions. The impact of particular noise types on facial identification is examined in this study using the Dark Face dataset to mimic real-world low-light situations. By examining brightness, pixel patterns, and image variations, noise was analyzed using a Python script. We tested four methods to reduce noise: Median Filtering, Non-Local Means (NLM), Wavelet Transform, and a deep learning model called DnCNN. These techniques were compared using quality measures like PSNR, SSIM, and MSE. Based on our research, median filtering works well for noises like salt-and-pepper noise but struggles to preserve facial features. On the other hand, DnCNN consistently performs better, enhancing image clarity and face recognition. This research offers important insights to increase the effectiveness of security and surveillance systems, particularly those that use facial recognition.