In recent years, Convolutional Neural Networks (CNNs) have promoted greatly the development of image retrieval, intelligent image retrieval still faces challenges. An intrinsic challenge in intelligent image retrieval exists the intention gap between the real intention of the users and the representation of users’ query, besides the well-known semantic gap. To address these problems, we propose a novel method that incorporates a relevance feedback (RF) method with an evolutionary stochastic algorithm, called multi-swarm of particle swarm optimization (MPSO), as a way to grasp the users’ perception of relevance through optimized iterative learning. One main component of our method, MPSO, can effectively prevent the retrieval system from falling into local optimal and dispose of those redundant particles, which can improve the diversity of particles. Moreover, we also present a simple but effective similarity ranking algorithm to increase retrieval speed, which can consider synthetically not only the fitness of each query point in feature space, but also the similarity of the image sequence corresponding to each query point. Extensive experiments on three publicly available datasets demonstrate that our method significantly improves the precision, recall as well as the user experience.
CITATION STYLE
Zhu, Y., Chen, Y., Han, W., Huang, Q., & Wen, Z. (2019). Intelligent Image Retrieval Based on Multi-swarm of Particle Swarm Optimization and Relevance Feedback. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 11954 LNCS, pp. 566–578). Springer. https://doi.org/10.1007/978-3-030-36711-4_48
Mendeley helps you to discover research relevant for your work.