Truy xuất hình ảnh sử dụng phương pháp kết hợp xếp hạng và Vision transformer

Abstract

Image retrieval is an important task in computer vision that involves retrieving relevant images from image databases based on their visual content. Traditional methods may not capture the semantic meaning of images effectively, leading to suboptimal retrieval results. In this paper, we propose a Content-Based Images Retrieval (CBIR) method based on Vision Transformer network architecture combined with the VisualRank method to rank images based on their similarity of images with query images. Through the training results, the proposed model has an accuracy of up to 97%.