The Research Of The Gray-Scale Image Segmentation Based On Improved Particle Swarm Optimization

Now image processing has been applied to many departments, such as military weapon systems and medical imaging systems, so image processing is becoming more important. Image segmentation technology is one of many key technologies in image processing, and started in the seventies of the last century, so far has decades of history. At present, many image segmentation algorithms address the issue of bi-level thresholding, only a few take into account the multilevel thresholding.The present paper proposed an intelligent multilevel thresholding segmentation algorithm for gray-scale image. The core of the algorithm is improved particle swarm optimization algorithm, which is FCLPSO-3. Here FCLPSO is the abbreviation of the Flexible Comprehensive Learning Particle Swarm Optimization. Particle Swarm optimization algorithm is a new category of intelligent algorithm, and it draws its inspiration from real-life swarm behavior, such as the processing of bird-Hocking.The velocity update rules of particles for FCLPSO-3algorithms are flexible, at the beginning of the algorithm, the new velocity of particles depends on the old velocity of particles, their current best position, and the global best position of the particle swarm, at this point the particle swarm quickly converges to the global optimization. In the middle of the algorithm, the new velocity of particles depends on their old velocity, their current best position, and the best position of the neighborhood of a certain range of the particles, now the particle swarm is divided into many neighborhood areas, in which all particles are flying to the best position of the current region, so the convergent rate of the whole swarm is slow, and the diversity of the population of the particle swarm is preserved well. At the end of the algorithm, the new velocity of particles depends on their current speed and the current best position of the sample particle, at the moment all particles move in the space, whose center is the current position of the particle, which can help the particle swarm to get away from the pseudo local optimization.The update rules of the velocity and position of the GCPSO algorithm are adopted by FCLPSO-3algorithm except the end and the main structure is the cooperative learning particle swarm optimization. In addition, a random particle is added to FCLPSO-3algorithm, the random particle is a particle except the best particle in the particle swarm, and its new position can be anywhere within the space of the particle swarm. FCLPSO-3algorithm can guarantee to converge to the global best position when the number of iterations tends to infinity. Finally, the simulation experiments on two classic gray-scale image is done, and the simulation results show that the segmentation results of the improved algorithm are very good.