AI Summary & Key Takeaways
AI InsightOverall, none of the studied CBIR descriptors can be used individually to build a full image retrievalsystem.
📌 3 Key Takeaways
Abstract
In recent years, multimedia searching has become an important research field. Multimedia filesare one of the most important materials on the internet. Unfortunately, even for the state-of-the-art methods andapplications based on accessing multimedia on the internet, it is hard to find the required files. The main purposeof this study is to investigate the performance of well-known image content-based retrieval techniques, i.e., FuzzyColor and Texture Histogram (FCTH), Edge Histogram Descriptor (EHD), Scalable Color Descriptor (SCD), ColorLayout Descriptor (CLD), Color and Edge Directivity Descriptor (CEDD), and Speed-Up Robust Feature (SURF)combined with Fast Library Approximate Nearest Neighbor (FLANN). In general, the objective of using thesetechniques is to find the query’s most relevant files and list them at the top of the retrieval list.Several experiments have been conducted and it has been observed that FCTH and SCD outperform other studiedtechniques. On the other hand, for the SURF combined with FLANN approach, the results of most of the querieswere below user expectations. In addition, extracting the feature vectors using this method requires massive amountof memory. Overall, none of the studied CBIR descriptors can be used individually to build a full image retrievalsystem. In our opinion, multiple descriptors can be used simultaneously to achieve a more robust system andaccurate results.