Performance comparison of DFT and Hartley plane sectorization in CBIR

This paper compares the retrieval performance of the CBIR system. In this system image features are extracted by means of Hartley and DFT transformed plane sectorization. Two different similarity measures named as sum of Absolute difference and the Euclidean distance are used for the matching the input image feature vectors with the feature vectors of the image database. The performances of the algorithm for both approaches, on different number of sectors are compared. The average precision-recall cross over point plot (PRCP) and the overall average PRCP performances have been compared for all sectors formed. This experiment makes use of the augmented Wang´s image database. This database consists of 1056 images belonging to 12 different classes. It has been observed that the best overall average PRCP performance is achieved when the Hartley transformed image plane is sectored into 4 sectors and the AD is used as similarity measure.