Improved Particle Swarm Optimization And Its Application In Adaptive Image Denoising In Shearlet Domain

Particle swarm optimization (PSO), developed from the research of the foragingbehavior of birds, is a swarm intelligence based random search algorithm. The PSOalgorithm is simple in concept and has few parameters, fast calculation speed and goodrobustness; it has gained great development in just a dozen years, has made goodapplications in some areas, and has become a new research focus.Although the PSO algorithm has good effect for simple optimization problems, inthe optimization of complex high-dimensional problems with multiple extreme points,the algorithm is prone to "premature" convergence in local optima. Therefore, how toavoid particles falling into local optima is an important research subject of thealgorithm.This paper first analyzes the principles, process and parameters of particle swarmoptimization algorithm, and introduces several common improved PSO algorithms,then introduces image noise model, image quality assessment, commonly useddenoising methods and Shearlet transform, and finally proposes several improvedparticle swarm optimization algorithms, and applys them to Shearlet image denoising.Detailed innovation points and research works are as follows:(1) To prevent particle "premature" convergence, using multi-model ideas, thispaper proposes a multi-stage multi-model improved particle swarm optimizationalgorithm. The algorithm divides the optimization process into three stages, each stageusing an evolutionary model with different exploration and exploitation capabilities,and to ensure population diversity and local optima refinement, the last two stagesiterate for a certain number of times, the last two stages operations are repeated untilthe optimal solution is found. Simulation results show that: the algorithm is more likelyto find the global optimum and is more efficient than two sub-swarms particle swarmoptimization algorithm.(2) By studying two sub-swarms particle swarm optimization algorithm (TS-PSO)and two sub-swarms optimization algorithm based on different evolutionary models(TSE-PSO), this paper proposes an improved two sub-swarms exchange particleswarm optimization algorithm (MTSE-PSO). Each sub-swarm in the algorithm uses adifferent model to evolve, and when the sub-swarm states are stable, a random part of particles from the first sub-swarm is selected to exchange with a part of pariticles withthe worst fitness values from the second sub-swarm, the above-mentioned operationsare repeated until the optimal solution is found. Simulation results show that:MTSE-PSO has better global search capability and convergence rate than PSO andTSE-PSO algorithms.(3) Because traditional threshold selection strategies “overkill” image factors andlose image details, this paper proposes a Shearlet adaptive image denoising methodbased on particle swarm optimization algorithm. According to the distributioncharacteristics of Shearlet transform domain coefficients of different scales anddirections, the method uses the multi-stage multi-model improved particle swarmoptimization algorithm to adaptively determine the optimal thresholds of differentscales and directions, to achieve image content-based adaptive denoising. Experimentresults show that the algorithm has better visual effect and higher PSNR values inimage denoising than standard particle swarm optimization algorithm.