40 likes | 169 Vues
This laboratory exercise focuses on implementing a Content-Based Image Retrieval (CBIR) system using MPEG-7 features. Participants will conduct experiments on weighted retrieval and approximate retrieval, utilizing the Java CBIR module based on the LIRE open-source library (version 0.9.4). Testing will be conducted on the UCID V2 dataset, with an emphasis on optimizing feature combinations such as CLD, SCD, and EHD for improved performance. Each group will summarize their findings using the provided Word and LaTeX templates and evaluate the results using MAP and P@10 metrics against the ground truth data.
E N D
Laboratory IMPEG7 implementation exercises Ing. Marco Bertini
Exercise objectives • Perform experiments with a sample CBIR system based on MPEG-7 features • Experiment with weighted retrieval • Experiment with approximate retrieval
Working tools • Use Java CBIR module based on LIRE open source library • use version 0.9.4 • Use UCID V2 dataset for testing. • Source code available for inspection • Word and Latex template for 4-page resume of the experiments (to be done per group)
Task • Using the CBIR module perform image retrieval by applying: • CLD • SCD • EHD and their (weighted) combinations. • Check the most appropriate feature combinations to optimize performance • Test on UCID dataset and evaluate MAP P@10 using ground truth data.