Abstract:
When searching a large database of images for ones that are visuallysimilar to reference images, it is desirable to have efficient means for making coarsecomparisons between images. In this project, we propose an approach for efficientimage retrieval using the Prometheus Orthonormal Set (PONS). In the proposedmethod, images are stored in the database in the form of PONS transformcoefficients of block decomposition. The energy spreading properties of this PONStransform suggest that global comparison features of images can be accomplishedusing any subset of the PONS transform' coefficients. We use the correlationcoefficient comparison to indicate the similarity between query image and retrievedimages. In addition, recall and precision measures are used to evaluate the system'sperformance. The 60.61 percent recall and 67.08 percent precision are obtained foroverall by using the proposed method. The 92.86 percent recall and 87.20 percentprecision are obtained for retrieving noised and blurred images.