Content-Based Image Retrieval: Impact of image resolution on the search accuracy and results ordering
Miroslav Marinov, Yordan Kalmukov, Irena Valova · 2021 International Conference Automatics and Informatics (ICAI) · 2021
Image resolution plays an important role in content-based image retrieval (CBIR) systems since both the query and the images being added to the database are processed pixel by pixel. Higher resolutions need more processing time, where the dependence is not linear, but quadratic. However, it is uncertain if high resolutions actually lead to more accurate search and better order of retrieved results. The purpose of this research is to determine the impact of image resolution on the search accuracy and results ordering. If it has a little influence on them, then it is quite appropriate to resize images before adding them to the image database. That could significantly speed up the insertion process. To test the impact of image resolution, we conduct a series of experiments using the same image dataset but pre-replicated in resolutions starting from 32x32 px to the images’ original resolutions, sometimes higher than 5000x3000 px. Results show that resolution of 128x128 px provides quite similar order to the reference one, having a correlation of about 0.97 between the two orderings. So it is worthy to resize all high resolution images (Full HD and especially 4K, 8K and higher) to 128 px in order to reduce the processing time while keeping the search accuracy and the order of results almost the same.