2023, Vol.26, No.4, pp.366 - 384
The paper describes the improvement of Content-Based Image Retrieval (CBIR) system efficiency. In CBIR, where each image is typically represented by low-level features (description of color, texture, and shape). It is proposed to solve the following two limitations: (1) not only exploiting the basic similarity measure of images, the paper proposed to use an improved Efficient Manifold Ranking (EMR), where the improved Fuzzy C-Means (FCM) clustering algorithm is used to determine the anchor points of the EMR Graph and (2) EMR has not yet taken advantage of both low- and high-level features of the image, it is the second limitation. To solve this limitation, the cf-EMR method is chosen to combine low-level features and CNN, in which the image is ranked with improved EMR to get the outstanding advantages of both features types, thereby improving the quality of access. So that the accuracy of image retrieval systems to be improved. The average accuracy of more than 85% of the tests, performed on the three image datasets Logo2K+, VGGFACE2-S and Corel30K, indicates the high efficiency of our recommendations in improving the performance of the CBIR system.
Key words: bigdata, content-based image retrieval (CBIR), Efficient Manifold Ranking (EMR), Fuzzy C-Means (FCM), Convolutional Neural Network (CNN)
DOI: https://doi.org/10.5281/zenodo.10410163
Full text: Acrobat PDF (2176 KB) Open Access
Copyright © Nonlinear Phenomena in Complex Systems. Last updated: December 21, 2023