Content Based retinal Image Retrieval scheme using Harmony Search Algorithm

Automated segmentation of vasculatures in retinal images is vital for the detection of Diabetic Retinopathy (DR). An attempt has been made to generate continuous vasculature information using evolutionary based Harmony Search Algorithm (HSA) combined with conventional Multilevel Thresholding (MLT) methods. The preprocessed normal and abnormal retinal images are segmented using HSA based Otsu and Kapur MLT methods by the best objective functions. The segmentation is validated with corresponding ground truth images using binary similarity measures. The statistical, textural and structural features are obtained from the segmented images and are analyzed. Content Based Image Retrieval (CBIR) is used to assist physicians in clinical diagnoses and research fields. The CBIR systems are developed based on both the MLT segmentation techniques and the obtained features. Similarity matching is carried out between the features of query and database images using the Euclidean Distance measure. Similar images are ranked and retrieved. This work shows high retrieval performances such as precision (96%) and recall (58%) for the CBIR system using HSA based Otsu MLT segmentation method than the other method. Hence this CBIR system could be recommended in computer assisted diagnosis for a better screening of the diabetic retinopathy.