Image Stitching based on Particle Swarm and Maximum Mutual Information Algorithm

Abstract

As a key link in image stitching, image registration based on maximum mutual information does not need any preprocess and has a high degree of automation and high registration accuracy, which thus attracted widespread attention. Optimization search in image registration process is easy to fall into local minima leading to the wrong registration parameters. This paper presents a method to build image pyramid based on wavelet transformation and uses swarm intelligence classical particle swarm optimization method to adjust its parameters with the mutual information and multi-resolution series. Experiments show that this method can effectively avoid local minima and can find the optimal registration transformation by finite optimization iterations. Performance of this proposed algorithm is better than the traditional multi-resolution method and its computational efficiency is higher than that of simulated annealing optimization method

Modified Sub-aperture Stitching Algorithm using Image Sharpening and Particle Swarm Optimization

“…A different optimization algorithm should be used to solve this problem. Some researchers have used particle swarm optimization to address this problem in image stitching [9]. Particle swarm optimization can be more reliable than the leastsquares method for finding the globally optimal solution.…”

Novel and robust methods for the automatic registration of image data

“…These methods introduce stochasticity for the creation of candidate solutions to our problem, which facilitates the escape from local optima. The most common ones that have been used are stochastic annealing [56,57], evolutionary optimization methods (Genetic algorithms [58,59], Evolution strategy [60,61], Differential Evolution [62,63])) and swarm optimization methods (Particle Swarm Optimization [64][65][66][67] , Artificial Bee Colony [68]). In order to escape local optima and find the global one, they have to do repetitive estimations of the similarity of the images for all the candidate solutions they generate.…”

“…PSO is combined with nonlinear least squares. [149] Rigid Quantum-behaved PSO with disturbance implementation strategy in order to increase diversity in the population and thus avoid entrapment in local minima [65] Affine Multi-resolution PSO [150] Affine Hybrid PSO which incorporates two concepts of GA, crossover and subpopulation [151] Rigid PSO with subpopulation and crossover [152] This variant combines Quantum PSO [153] with [66] Rigid The velocity of each particle x i is calculated as…”