Research And Application Of Image Segmentation Based On High-order MRF Model

In recent years,image segmentation is a hot research topic in the field of computer vision and image processing.The quality of image segmentation results has an important impact on the application of computer vision,such as scene understanding,target tracking,target recognition and other applications.For complex natural images,existing image segmentation algorithms are difficult to obtain the expected results.Especially in actual scenes,natural images often have high noise and uneven illumination,which greatly increases the segmentation of natural images in actual scenes.At present,in the field of image segmentation,the image segmentation method based on Markov random field model(MRF)is a research hot research topic.This research is aimed at the characteristics of complex natural images and provides an effective solution for image segmentation in natural scenes.This scheme is based on the high-order Markov model,combined with the asymmetric Gaussian mixture model,introduces edge constraint terms,and proposes an improved image segmentation method based on high-order MRF model.The main contents of this research include the following aspects:This research takes complex natural images as the research object.Aiming at the problems of natural image including high noise,texture confusion and uneven illumination,many image segmentation algorithms have been proposed at domestic and foreign.The research status of probability graph model,Markov random field model and Gaussian mixture model in image segmentation are introduced.The image segmentation method based on high-order Markov random field model is studied.The image segmentation method based on high-order Markov random field model is researched,that compared with the traditional image segmentation model.(2)The problem that the traditional MRF model is sensitive to image noise in segmenting complex natural images.Based on the research of Pairwise MRF model,a high-order MRF(RLSI-HMRF)image segmentation model based on robust local spatial information is proposed.Firstly,the local spatial relationship is measured based on Hamming distance,and the weighted Gaussian mixture model(WGMM)is constructed according to its similarity.The statistical features of the local region of the image are described,and it is established that the likelihood feature of the local spatial class consistency from the label space to the pixel intensity field.Secondly,based on the Robust model,the spatial global constraint relationship of long distance is introduced,and the robust high-order MRF regional label consistency constraint is established.Finally,the RLSI-HMRF image segmentation model is established under the MRF model framework.Experiments show that the proposed model has an improved image segmentation effect.(3)The traditional GMM describes the distribution of complex natural image feature information data not accurate,make the results of image segmentation inaccurate.Based on the research of RLSI-HMRF image segmentation model,this research proposes a model of high-order MRF image segmentation model based on asymmetric Gaussian mixture model(AGMM-HMRF).An asymmetric Gaussian mixture model is introduced based on the Gaussian mixture model.Firstly,an asymmetric Gaussian mixture likelihood feature is established to describe the local spatial class consistency from the observation field to the label field.Secondly,the asymmetric Gaussian mixture likelihood feature of local spatial class consistency is introduced from the framework of RLSI-HMRF image segmentation model.A high-order MRF image segmentation model based on asymmetric hybrid Gaussian is established.The experimental results show that the proposed model can better describe the natural image information and get a good segmentation effect.(4)For the traditional MRF model,when the road scene image is segmented,the road structure is unclear and the image edge is blurred.Based on the research of AGMMHMRF image segmentation model,it is proposed that a high-order MRF image segmentation model based on asymmetric Gaussian mixture model with edge constraints(EAGMM-HMRF).It is introduced that based on the AGMM-HMRF image segmentation model,a structural forest edge detection algorithm.The edge information of the road scene image is extracted and the edge constraint of the image segmentation model is established.Under the framework of the AGMM-HMRF model,the local and regional features of the image are fused.It establishes an EAGMM-HMRF image segmentation model.The experimental results show that the proposed model can better maintain the edge structure information of the road scene image and has better segmentation results.Finally,the comprehensive research content summarizes the research work and research results.And this thesis analyzes the problems found in the research and the shortcomings.At the same time,we will look forward to the future development of the research content and further research directions.