Combining EMR rankings for multi-image queries

In CBIR (Content-Based Image Retrieval), using a single query often leads to limitations in semantic image search. The Efficient Manifold Ranking (EMR) algorithm has been widely used in CBIR and often uses low-level or high-level features for image querying. However, high-level features from Convolutional Neural Networks (CNNs) have more advantages, helping to describe the semantics of images more clearly than low-level features such as color, texture, and shape.To improve the performance of the EMR-based image retrieval algorithm in content-based image querying, this study proposes a method to combine the rankings of each query image. Instead of relying on only a single query image, we combine the EMR ranking results of two input query images into a single ranking. This method helps to increase the accuracy and efficiency of querying by leveraging information from different perspectives of semantics.Experiments were conducted on the VGG60K dataset to demonstrate the effectiveness of the proposed ranking method in improving the quality of image retrieval.