Deep Learning-Assisted Localized Content-based Image Retrieval Framework with Multiple-Instance Learning Algorithm

Parvathy Arulmozhi

doi:10.1142/S0218001425520196

Return

Deep Learning-Assisted Localized Content-based Image Retrieval Framework with Multiple-Instance Learning Algorithm

2025-09-01

PRE

Parvathy Arulmozhi

R. Gopi *

DOI：10.1142/S0218001425520196

Original

Original request for help

Save

Abstract

En 中文

Traditional content-based image retrieval (CBIR) searches a database for pictures that match a single, unannotated query image. A comprehensive (or worldwide) perspective of the picture is essential for this kind of search. However, the intended picture content is frequently not global rather regional. Computer vision, pattern recognition, and CBIR are just a few of the applications that have long faced the semantic gap issue: how to bridge the gap between human perception of low-level semantic concepts and machine collection of high-level image pixels. In light of the recent achievements in deep learning research, there is hope for bridging the semantic gap. Therefore, the deep learning-assisted localized content-based image retrieval framework (DL-LCBIRF) was suggested in this study to rank photos in the database according to a similarity metric dependent upon specific areas inside the image. The first layer consists of generic descriptors that stand in for groups of feature vectors that are comparable and rotationally invariant. The combined probability of the frequencies of the generic descriptions over neighborhoods makes up the second layer. The proposed DL-LCBIR uses labelled images combined with a multiple-instance learning algorithm (MILA) to locate the target item and adjust the feature weights. This multi-modal probability is shown as a collection of spatial frequency clusters. It augments rotationally invariant statistical spatial constraints. In addition to enhancing the model's performance, choosing a unique structure determines the model's distinguishing features, which are common in good cases and seldom in negative ones.

Keywords:

Deep learning

content-based image retrieval

multiple-instance learning algorithm

medical content-based image retrieval

neural network