New CLAHE pipeline boosts retinal image clarity by 37% for earlier disease diagnosis

In a new paper published on arXiv (2607.17691), researchers K. Mithra and Prem Kumar Santhanam introduce a luminosity-adaptive contrast enhancement method for retinal fundus images. Retinal imaging is critical for diagnosing sight-threatening conditions like diabetic retinopathy, glaucoma, and retinal vein occlusion, but uneven illumination, motion blur, and low contrast often compromise image quality. The proposed pipeline first corrects luminosity by decomposing the image into HSV color space, then applies Contrast Limited Adaptive Histogram Equalization (CLAHE) exclusively to the Value channel. This preserves color integrity while enhancing contrast.

Experiments on the public DRIVE dataset (40 fundus images, 584×565 pixels, Canon CR5 camera) show the method achieves a Peak Signal-to-Noise Ratio (PSNR) of 29.3 dB, Structural Similarity Index (SSIM) of 0.91, and Contrast-to-Noise Ratio (CNR) of 3.12. These results significantly beat standard Histogram Equalization (PSNR=21.4 dB, SSIM=0.74) and Adaptive Histogram Equalization (PSNR=23.1 dB, SSIM=0.79). The pipeline processes each image in an average of 0.14 seconds, fast enough for real-time clinical screening.

A subsequent binary masking step isolates hyper-reflective regions consistent with vascular pathology, further aiding computer-aided diagnosis. The authors note limitations and suggest future comparisons with deep-learning-based methods. Overall, this approach offers a computationally efficient, high-fidelity enhancement that could reduce diagnostic errors in ophthalmology.