Research And Application Of Image Segmentation Based On Improved FCM Algorithm

Image is a similar and vivid description to the things in objective reality. It is also an important measurement to obtain information, expression information and deliver information.There is about seventy-five percent of information presenting in front of people as the form of images. Therefore, it needs to segment critical sections from image to effectively use the information, and then the target can be measured and analyzed, etc. It is thus clear that image segmentation is the foundation of image processing. Segmentation results will directly influence the further operations such as image analysis and image understanding. As a result,image segmentation has always been a focus and difficulty in research of image processing.At present, there does not have an image segmentation method suitable for all images,because different fields have different image features. So the researchers propose different type of image segmentation methods as far as possible to satisfy the actual division requirements. Currently, the improved image segmentation strategies can be divided into two aspects. On the one hand, starting from the algorithm itself, mathematical expressions are changed or new constraint conditions are added to update the objective function. On the other hand, with the help of new theories and new methods, many existing image segmentation methods are combined with them to produce new image segmentation method. The fuzzy C-means(FCM) clustering algorithm is the typical representative. Because it can solve the uncertainty of image pixels very well, so it gets widespread attention and study. This thesis makes a deep research on the FCM algorithm, and then puts forward improved strategies through analyzing disadvantages of the FCM algorithm. The main research contents of this thesis are as follows.Firstly, because FCM algorithm is sensitive to the initial clustering center, so quantum particle swarm optimization algorithm is used to seek out the optimal solution, and then the best solution is decoded as the initial parameters of FCM algorithm. Particle swarm optimization algorithm has stronger global search ability; however, it is extreme falling into local minimum value, so quantum algorithm is introduced. Quantum revolving door is adopted to update the movement of the particles. Quantum NOT gate is adopted to increase the diversity of the population. Therefore, the local optimization ability of particle swarm optimization algorithm has been greatly improved, which enhances the global convergence ability of particle swarm optimization algorithm at the same time. Through experiment simulation, the improved algorithm has better segmentation effect than traditional FCM algorithm.Secondly, taking useless noise rejection capability of FCM algorithm into consideration,an adaptive weighted spatial information FCM algorithm is put forward to improve its abilityto resist noise. Adaptive weighted coefficient can distinguish the influence degree of the noise data to its center, thus it can reduce the noise influence. Therefore, introducing new space constraints, we can define the objective function which deduces the clustering center renewal expression and the membership matrix renewal expression. Through experiment simulation,the result shows that improved algorithm has a superior ability to different noises.Finally, for rail gap image having gray darker and easily involving the influence of noise,the improved FCM algorithm is applied to the rail gap image segmentation. First of all, the quantum particle swarm optimization algorithm is used to find the initial clustering center of FCM algorithm, and adaptive weighted space information FCM algorithm which has stronger noise resistance is used to finish rail gap image segmentation. Through experiment simulation,the improved FCM algorithm can segment relatively complete key components, which lays a solid foundation for the rail gap follow-up offset measurement.