A novel approach for fundus image enhancement

Fundus image enhancement is an essential and challenging pre-processing step for automated diagnosis of ocular disorders. The enhancement works reported earlier were developed for the distortions such as additive white Gaussian noise and salt-and-pepper noise. However, this poses a significant limitation for the applicability of these methods, as occurrences of such distortions are least likely. In this work, the five most common distortions are identified, and algorithms are proposed to create distortions resembling the same. Thereafter, a residual dense connection based UNet (RDC-UNet) architecture is proposed for the enhancement task. The residual dense connections incorporated in the UNet effectively captures both local and global information from the images beneficial for the enhancement task. The RDC-UNet was trained individually for each of the five distortions and then applied to the synthetic degraded fundus images. The experimental results show that the visual quality and quantitative results are, on average, 8% better than the state-of-the-art methods reported in the literature. Furthermore, in case of naturally degraded images, the type of distortion is not known apriori. Additionally, multiple such distortions can be present at a time. An ensemble model architecture is proposed using the RDC-UNet trained individually for each degradation to address this challenge. Experiments conducted over naturally degraded fundus images demonstrate that the proposed model effectively enhances the visual quality of fundus images. In addition, the effectiveness of the proposed method is also shown with the application of blood vessel segmentation.