An approach to content based retinal image retrieval using papamarkos multilevel thresholding method

In this work, a framework for automated content based image retrieval system is developed for the identification of diabetic retinopathy in retinal images. A multilevel threshold based Papamarkos algorithm is applied on preprocessed retinal images to segment the blood vessels and corresponding statistical texture and structural features are extracted. Quantitative analysis of derived features also performed. Similarity matching is carried out using Bhattacharyya distance measure. The performance validation of this content based image retrieval system is calculated in terms of retrieval measures. The Papamarkos multilevel thresholding algorithm is able to identify most of the blood vessels in normal images. The lesions in the abnormal conditions are also extracted. The derived features show better differentiation of normal and abnormal retinal images. The content based image retrieval system using this method performs better with high degree of precision and recall. It appears that the proposed method is able to retrieve similar images in large database. Hence the Papamarkos multilevel thresholding based CBIR system could be useful in automated screening of diabetic retinopathy.