Improving Person Re-Identification Performance Using ESRGAN Image Enhancement

Abstract

Person re-identification is one of the most important challenges in the field of image processing and computer vision, which has attracted significant attention in recent years. One of the main challenges in this domain is the low quality of images captured by different cameras, which can result in the loss of important details and consequently a reduction in the accuracy of re-identification algorithms. This paper presents a framework that enhances network accuracy and capability in determining individuals’ identities by improving the brightness of low-light images using histogram equalization techniques and increasing their resolution using the ESRGAN network. The experimental results show that the proposed method, by improving the brightness and resolution of images which plays a key role in revealing important details, enhances the quality of the re-identification network's input. Also, compared to existing methods, our model increases the accuracy of person re-identification, such that it achieves Rank-1 accuracy of 93.6% for the CUHK01 dataset and 94.5% for the CUHK03 dataset.