A powerful method for interactive content-based image retrieval by variable compressed convolutional info neural networks

There is a need for efficient methods to retrieve and obtain the visual data that a client need. New methods for content-based image retrieval (CBIR) have emerged due to recent developments in deep neural networks. However, there are still issues with deep neural networks in interactive CBIR systems like the search goal needs to be preset, scrambling and the computational cost is too high for an online environment. By this concern, this manuscript proposes an effective interactive CBIR that accurately retrieves images in response to the image query using variable compressed convolutional info neural networks (VCCINN). The weight of neural network is optimized by the variable info algorithm, and the matching activity is done by recursive density matching. The interactive technique eliminates irrelevant images based on user feedback and only the relevant images are finally retrieved. The overall retrieval performance in caltech-101 (dataset 1) and inria holiday (dataset 2) are 98.17% and 99% respectively. The performance of introduced model is proven by conducting ablation experiment on each component. The differential learning-based introduced image retrieval approach outperforms several existing methods regarding image similarity and retrieval speed.