Research On Image Segmentation Technology Based On Spectral Clustering

Computational vision is an interdisciplinary domain technology that realizes automatic machine recognition and image understanding by means of intelligent information processing,pattern recognition and the like.This technology has become a research and application hotspot in many fields,including: remote sensing image processing,industrial intelligent manufacturing(part defect detection),medical diagnosis,automatic driving and public opinion monitoring.Image segmentation is one of the key links in the field of visual computing,through high quality segmentation can accurately obtain user interested content,and make the following of image semantic understanding precisely.After years of research and development,image segmentation has formed a variety of effective segmentation methods,including Region-based segmentation technology,Edge-based segmentation technology and machine learning-based segmentation.Image segmentation technology based on machine learning has become the main research method in this field.Clustering is an unsupervised machine learning method,which relies on the designed rules to realize automatic data classification.Clustering technology has been widely used in the fields of sociology,medicine and computer vision.Developing new application fields of clustering is also one of the hot research directions in machine learning.Spectral clustering is a clustering algorithm based on graph theory,in recent years has been many researches and applications.The algorithm not only has sufficient theoretical basis,but also can better overcome the local optimum and achieve accurate solution to the problem.The basic principle of the general clustering algorithm is to project the original data into a feature space of low dimension,which can reduce the computational workload and improve the calculation speed.On the other hand,it can avoid the singularity caused by the direct recognition of the data with excessive dimensions.Problem to accurately identify samples of any shape.Benchmarking-based optimization(BBO)is a novel swarm intelligence algorithm that simulates the final solution of the problem by simulating the internal group of the company to learn from the company's benchmarking characters.This thesis mainly studies digital image segmentation based on spectral clustering.In order to reduce the amount of computation in image segmentation,a spectral clustering algorithm for image sample preprocessing is proposed.The k-means++ algorithm is usedto pre-segment the image first,and then clustering is implemented again by spectral clustering.Because the spectral clustering algorithm has strong sensitivity to parameters,in order to improve the image segmentation quality and improve the robustness of the algorithm,a multi-spectral clustering integration mechanism is designed to generate image segmentation results for spectral clustering of different parameters.Effective integration.In order to improve the integration effect of the final result,a reverse learning benchmarking optimization algorithm is proposed to achieve the optimal matching search for mutual information between integrated members.Finally,the effectiveness of the algorithm is verified by a series of experiments.