DCT-Based Downscaling Effects on Color and Texture-Based Image Retrieval

This paper evaluates the effects of downscaling in compressed domain on content-based image retrieval based on experimental studies. Several experimental databases are generated based on uncompressed ones using JPEG compression and DCT-scaling. Color and texture features are used in the retrieval queries, which constitute the base for the experiments. The query results are graded subjectively, and then the grades are used to obtain statistical values for evaluating the concrete effects. The evaluation results show that downscaling does not have significant impact on color-based retrieval in general, while it degrades texture-based retrieval performance significantly. The results also verify that querying uncompressed images in uncompressed databases gives the best color and texture-based retrieval performance. Querying compressed images in uncompressed and compressed databases yields similar color-based retrieval results for each scaling factor. On the other hand, querying compressed images in uncompressed databases drastically degrades the texture-based retrieval performance.