Research On Image Segmentation Algorithm Of FCM Based On Superpixel

Image segmentation is one of the most challenging research topics in computer vision and pattern recognition.It is the process of dividing the image into several regions with unique properties.The quality of segmentation affects the effect of subsequent processing.With the development of research,scholars have proposed a large number of image segmentation algorithms.Among them,the image segmentation algorithm based on fuzzy clustering is the most classic.It combines fuzzy properties and clustering theory to express better results,and has been widely used in many research fields.Because Fuzzy C-mean clustering(FCM)algorithm has repeated distance calculation in the iteration of clustering center and local neighborhood information,it leads to high complexity in the merging calculation of local spatial information and long time consuming of the algorithm.The lack of spatial neighborhood information of pixel and membership degree,and the regularization of neighborhood window of pixel can not express the real local spatial structure of image,resulting in poor segmentation effect.To solve the above problems,scholars put forward a variety of solutions.This thesis focuses on the improvement idea of FCM and does the following work:Firstly,the theoretical basis of standard FCM algorithm is introduced,the advantages and disadvantages of FCM algorithm and fuzzy clustering are analyzed,and the improved FCM algorithm in recent years is introduced emphatively.By summarizing the algorithm,this thesis makes an in-depth study on FCM image segmentation algorithm based on superpixel.Secondly,introduces the advantages and disadvantages of the commonly used color space and super pixel image acquisition method,aimed at the disadvantages of its watershed is pixel algorithm,through the adaptive morphological gradient image reconstruction to overcome the over-segmentation of its improvement,while removing useless minimum reserve the detail of the meaningful,precisely in order to obtain contour pixels in the image.Using it as a preprocessing step of FCM algorithm not only provides adaptive local spatial neighborhood,but also simplifies the original image.Finally,in view of the FCM algorithm using Euclidean distance to measure similarity between super pixel faults,introduces the ultra pixels in its objective function image color information,and can improve the classification of high-dimensional data hidden markov random field,will be better able to measure the similarity between the clustering center and the pixel area,as a fuzzy measure criterion so as to improve the effect of clustering.In this thesis,the experimental results of real images and composite images show that the improved algorithm can improve the clustering effect while retaining the target details,and it has a good anti-noise performance in the composite images with added noise,and can complete the segmentation in a very short time.In conclusion,the algorithm achieves the desired results in segmentation effect,anti-noise performance and running time.