Detail-richest-channel based enhancement for retinal image and beyond

摘要

The quality of retinal photography is not always satisfactory due to the limited imaging conditions. Low contrast and insufficient brightness are common problems. By comprehensively analyzing the image formation model from an intuitive perspective, we present general principles of parameter settings for different enhancement scenarios, and this analysis provides theoretical support for designing the proposed method. In a retinal image, the green channel preserves the richest details due to the red free photography. We term the green channel as the "detail-richest-channel" in a retinal image, and it is used as the initial transmission map. Subsequently, the transmission map is corrected by the modified cubic function to reduce over enhancement. The proposed method is designed for enhancing retinal images photographed by handheld fundus camera initially, and it can be extended to other applications. A total of 1045 retinal images photographed by handheld fundus camera, 460 retinal images acquired from cataract patients using high-end fundus camera, and 890 underwater images are tested to validate the effectiveness of the proposed method. Results show that the proposed method can improve the quality of degraded images as well as preserve the natural visual perception.