Particle Swarm Programming-Based Interactive Content-Based Image Retrieval

Particle structure in particle swarm optimization (PSO) is fixed in initialization and may result in premature or slow convergence. To tackle this problem, an improved PSO approach called particle swarm programming (PSP) is presented. PSP forms flexible nonlinear distribution representation of particles by introducing hierarchical tree structure into PSO. Furthermore, PSP is introduced in relevance feedback (RF) process of interactive content-based image retrieval (CBIR) by constructing a nonlinear updated query vector. Tests with five benchmark functions demonstrate that PSP can indeed increase diversity of initial particles, enhance search power and improve convergence over PSO. Extensive experiments on Corel-1000 and Catch-256 datasets show that the proposed PSP-based CBIR technique outperforms other linear or recent RF methods proposed for CBIR.