Content Based Image Retrieval

This project provides a basis for comparison of the retrieval performance of the CBIR system. Two similarity measures namely sum of Absolute difference and Euclidean distance are used along with, Precision and Recall. The performances of the algorithm for all the different approaches on different sector sizes are compared. The average and overall average precision-recall crossover point performances have been compared over 4, 8, 12 and 16 Sectors. This experiment makes use of the augmented image database. It uses the sectorization method on DFT and Hartley transforms . A comparison and evaluation of these methods is carried out and the results are drawn based on the retrieval rates of each class of images from the database. Three different similarity measures such as Precision, Recall, Precision- Recall Crossover are used for the matching the feature vectors of the query image with that of the images in the database. Euclidean distance and sum of Absolute difference are used for measuring the distances between the images. The average precision-recall cross-over point plot (PRCP) and the overall average PRCP performances have been compared for all the methods.