This research explores the impact of photo distortion on face recognition algorithms performance, focusing on convolutional neural networks (CNNs) due to their superior efficiency. Through a comprehensive review of current solutions and empirical experimentation with the LFW image database, this study investigates how compression, blurring, and noise affect algorithmic accuracy. Experiments utilized the Python-based DeepFace library for comparative analysis of photo similarity, optimizing algorithm parameters through density analysis and decision trees. The study confirms that while compression minimally affects performance, noise—especially at higher levels—significantly degrades accuracy. However, applying median filters to noisy images largely restores effectiveness. Blurring’s impact varies with direction, with vertical blurring severely impairing recognition capabilities. Among the algorithms evaluated, FaceNet512 emerged as the most resilient in handling various types of distortions. The findings underscore the importance of distortion type and parameter optimization in maintaining recognition accuracy, offering valuable insights for enhancing algorithmic resilience against photo quality degradation.