Early identification of diabetic retinopathy is highly beneficial for preventing the progression of disease. Appearance of blood vessels & retinal surface is a good ophthalmological sign of diabetic retinopathy in fundus images. In this paper, a novel method involving two approaches has been proposed for diagnosis of diabetic retinopathy. The first approach deals with estimation of fractal dimension of lesions by applying power spectral fractal dimension algorithms. For healthy retinas, fractal dimensions are found to be in the range of 2.00 to 2.069, whereas for retinas with diabetic retinopathy, fractal dimensions exceed upper limit. In the second approach, Gray Level Co-occurrence Matrix method is used to analyze the extracted regions from healthy and diabetes affected fundus retinal images. Texture features such as entropy & contrast are computed for healthy and unhealthy regions. These texture features are compared with fractal dimensions. The authors observed positive correlation between entropy and fractal dimensions, whereas negative correlation with contrast and fractal dimensions. Detailed implementations of the proposed work are presented.

No abstract