Analysis of Human Retinal Vasculature for Content Based Image Retrieval Applications

摘要

In this work, an attempt has been made to analyse retinal images for Content Based Image Retrieval (CBIR) application. Canny edge based CBIR systems are developed with and without preprocessing techniques. Blood vessels of normal and abnormal retinal images are segmented using Canny edge method. The structural and texture based features are obtained from segmented images. Similarity comparison is carried out using Bhattacharyya distance measure. The retrieved images are ranked. Retrieval efficiency of the CBIR systems is compared based on their performance measures such as precision and recall. The results demonstrate that features derived using Canny with morphological preprocessing could differentiate normal and abnormal retinal images significantly. Precision and recall of the CBIR system using Canny with preprocessing is found to be better than without preprocessing. It appears that this CBIR system aids in diagnosis of retinal abnormalities.