Optimized image enhancement of colour processing for retinal fundus image

Identifying Diabetic Retinopathy (DR) features based on retinal fundus images traditionally involves eye examinations by ophthalmologists. However, these original retinal fundus images often exhibit challenges such as low contrast, non-uniform illumination, and colour inconsistency, which can impact the accuracy of DR classification. Therefore, enhancing image quality by applying colour image processing techniques to the original retinal fundus images is crucial. This study introduces two novel techniques designed to overcome the limitations of existing algorithms. Firstly, a new colour correction algorithm named Fuzzy Tuned Brightness Controlled Single-Scale Retinex Histogram Matching (fTBCSSRhm) is proposed to address the issue of colour inconsistency in the dataset. Secondly, the enhanced Tuned Brightness Controlled Single-Scale Retinex with Hybrid Particle Swarm Optimization - Contrast stretching (eTBCSSR-HPSOCS) algorithm is introduced to tackle the limitations of the standard Particle Swarm Optimization (PSO) algorithm in HPSOCS, which is prone to local optima and exhibits low convergence rates. This technique combines the L-component of the LAB colour model with an enhanced velocity mechanism in PSO and contrast stretching (lavHPSOCS). Its goal is to fine-tune parameters automatically, reduce over-enhancement, avoid unwanted artifacts, and preserve intricate details. This approach aims to improve optimization by balancing exploration and exploitation and refining velocity control. The proposed algorithm underwent both qualitative and quantitative evaluations. Tests on 600 retinal fundus images from primary and secondary datasets were performed to benchmark the algorithm two existing approaches. The results show that the qualitative performance of the proposed enhancement is more favourable to ophthalmologist specialists compared to other images. Quantitatively, the method outperforms others with the lowest mean squared error (MSE) of 44.729, the highest peak signal-to-noise ratio (PSNR) of 32.768, and entropy of 0.977. It achieved a 95.034% success rate in classification accuracy. The study introduces a new colour correction model and an optimized image enhancement model, significantly improving retinal fundus image quality and establishing the most effective model for image quality enhancement