HDR-Like Image Generation to Mitigate Adverse Wound Illumination Using Deep Bi-directional Retinex and Exposure Fusion

Periodic assessment is necessary to evaluate the healing prog-ress of chronic wounds. Image analyses using computer vision algorithms have recently emerged as a viable alternative that has been demonstrated by prior work. However, the performance of such image analysis methods degrade on captured in adverse illumination, which is common in many indoor environments. To mitigate these lighting problems, High Dynamic Range (HDR) image enhancement techniques can be used to mitigate over- and under-exposure issues and preserve the details of scenes captured in non-ideal illumination. In this paper, we address over- and under-exposure simultaneously using a deep learning-based bi-directional illumination enhancement network that is able to generate over- and under-exposed images that are then fused into a final image with enhanced illumination. In rigorous evaluations using metrics including structure similarity, peak signal-noise ratio and changes in segmentation accuracy, our proposed method outperformed the state-of-the-art (SSIM scores \(0.76\pm 0.04\)/\(0.69\pm 0.08\) on bright/dark images, PSNR scores \(28.60\pm 0.70\) on dark images and DSC scores \(0.76\pm 0.09\)/\(0.74\pm 0.09\) on bright/dark images).