Аннотация
This paper presents an analysis of the impact of low-light image enhancement on face detection and recognition performance. To evaluate this problem, The Extended Yale Face Database B which contains face images captured under various illumination conditions and pose variations was used. The comparison was conducted across three variants including original images as well as images processed using the Zero-DCE++ and EnlightenGAN methods. The evaluation was carried out sequentially and included an analysis of face detection results, a determination of the number of images suitable for subsequent recognition and the calculation of closed-set identification rates. For a more comprehensive comparison, two evaluation modes were used: one on the sets of acceptable images for each method and one on a common set of images present in all compared methods. The results showed that image enhancement increases the number of successfully detected faces and images suitable for recognition. It was also found that the difference between the compared methods persists when analyzed on a common set of images, indicating the impact of processing not only on the detection stage but also on the final identification rates. These findings may be useful in developing face recognition systems designed to operate in low-light conditions.