Swarm Features Applied in Deep Learning for CBIR of Chest X-Ray Images

Abstract

The chest image retrieval process involves searching entire images in database for accurate image retrieval. The image retrieval process consumes hardware resource and time depending on query image feature complexity, and database size. Traditional algorithms such as neural network and genetic algorithm consume more time for image retrieval due to hardware limitation and consumes more time for training neural network. In this paper, we propose Particle Swarm Optimization (PSO) and Capsulenet deep learning framework to minimize time consumption for Content Based Image Retrieval (CBIR) and improve precision, recall of retrieved images. The PSO-Capsulenet determines feature vectors of Chest X-ray image. The feature vectors of image obtain by evaluating color matrix histogram, shape and color features of X-ray images. The PSO-Capsulenet minimizes time consumed for image retrieval by 20% and improves retrieved image accuracy by 40% compared to traditional algorithms.