Document Type : Original Article

Authors

• Ali Sabzkar
• Mohammad Hossein Khosravi

Faculty of Electrical and Computer Engineering, University of Birjand, Birjand, Iran

Abstract

Reliable retinal image quality assessment (RIQA) is a critical prerequisite for accurate computer-aided diagnosis in ophthalmology, particularly in telemedicine and large-scale screening programs. This study proposes a no-reference retinal image quality assessment framework that evaluates structural degradation and contrast distortions using statistical image descriptors and machine learning classification. The proposed method extracts Mean Subtracted Contrast Normalized (MSCN) coefficients together with grayscale, color, and Color-Gray Difference (CGD) histogram features from image patches. Compact image representations are obtained by performing covariance eigen-decomposition independently on the MSCN and histogram feature matrices of each image and retaining the dominant eigenvalues as statistical descriptors. The resulting eigenvalue-based feature vectors are concatenated and classified using a Support Vector Machine (SVM) with a radial basis function (RBF) kernel. Experimental evaluation on the DRIMDB and ACRIMA databases achieved accuracies of 0.98 and 0.89, respectively. The proposed method demonstrated competitive performance on DRIMDB and achieved the best overall performance on ACRIMA compared with existing state-of-the-art approaches.

Keywords

• Retinal Image Quality Assessment
• MSCN coefficients
• Image Histogram
• Dimension reduction
• Structure degradation
• Contrast degradation