Determining an approach to iris recognition depending on shooting conditions

The object of this study is the development and evaluation of image processing and analysis methods for iris recognition, which can be integrated into human-machine interaction (HMI) systems based on biometric data or other contactless interaction approaches. Enabling high accuracy and reliability of biometric iris recognition systems under variable imaging conditions remains an open scientific challenge. One of the primary difficulties is the impact of changing lighting conditions, head tilt, and partial eye openness on identification results. This study assesses the effect of preprocessing methods (Equalization Histogram, CLAHE) on iris image quality and compares the algorithmic method (Hamming Distance) with neural network models (CNN, DenseNet) based on key metrics, including accuracy, False Match Rate, False Non-Match Rate, and Equal Error Rate. Additionally, the influence of training dataset structure and neural network hyperparameters on classification performance was analyzed. The results demonstrate that the Hamming Distance method (HD = 0.35) achieves 95.5 % accuracy, making it a competitive alternative to neural networks. It was established that combining CLAHE and Equalization Histogram effectively reduces noise and enhances segmentation accuracy. Furthermore, it was determined that the DenseNet-201 neural network achieves an accuracy of 99.93 % when using an optimal dataset split (70 %:15 %:15 %). The study confirms that preprocessing techniques such as normalization and adaptive contrast enhancement significantly reduce recognition errors under varying lighting conditions. The proposed solution holds significant potential for assistive technologies for individuals with visual impairments, the automotive industry, as well as security systems