DDBTC approach with binary particle swarm optimization for greedy-DCNN based CBIR system

Abstract

Recently, the researchers have been focused on making image retrieval as effective as possible since it has significant importance in many fields. A semantic distance between the features and the human concepts derived from images has become a major problem that reduces retrieval precision. Several algorithms have been created between the images to minimize this semantic gap. However, no successful algorithm for extracting images from the massive database has been proposed. As a result, retrieving data from a huge database array has become more difficult till now. So, in order to address these issues, a dot diffused block truncation coding (DDBTC) with meta-heuristic optimization named binary particle swarm optimization (BPSO) is introduced in this paper. Moreover, the novel optimization based DDBTC method to solve the proposed CBIR optimization task. The optimized DDBTC is utilized to develop the image feature descriptor such as Colour Histogram Features (CHF), Colour Co-occurrence Features (CCF), Bit Histogram Feature (BHF), and Bit Pattern Feature (BPF). Besides, to proficiently solve the dictionary learning issues in the compressed domain Greedy-DCNN based Dictionary Learning algorithm is introduced. The Euclidean distance metric measures the similarity among two images. The proposed scheme is implemented, and performance measures such as precision, accuracy, and recall are utilized to evaluate performance. The performance is compared to various feature vectors, and the introduced scheme's run time/retrieval time is compared to existing BTC methods. It shows the introduced scheme achieved better retrieval time than the ADBTC scheme. The accuracy achieved by the proposed approach for three different datasets are found to be 98% (corel 1 k dataset), 99% (corel 10 k dataset), and 96% (Caltech dataset).