Mixture Model With Spatial Information For Image Segmentation

With the continuous progress of artificial intelligence,deep learning and computer vision technology,digital image processing is applied to the most promising fields such as advanced manufacturing,unmanned aircraft and driverless cars.Therefore,the progress of image processing technology is the most urgent.Under this social background,we have carried on the research work on image decolorization and image segmentation.First of all,we studied the contrast preserving decolorization model,hoping that the converted grayscale images can reproduce the visual experience of color images and save more details and feature discriminability.These features can be used for image segmentation.Then,we mainly concentrate on the study of image segmentation technology.So far,researchers have made a lot of achievements in the field of image segmentation,which are derived from the use of different features in images.But so far no a segmentation algorithm can be used for all the image segmentation,it also promoted the researchers consistently research on image segmentation.The main research work of this paper mainly includes the following aspects:(1)Due to the loss of contrast in the current image decolorization model,this paper presents a new color-to-gray conversion model based on Log-Euclidean metric and Gaussian kernel function.Based on the Log-Euclidean geodesic metric of lie group and Riemannian manifold,we propose an efficient decolorization framework.In the proposed model,motivated by the fact that Log-Euclidean metric has promising invariance properties such as inversion invariant and similarity invariant,we present a Log-Euclidean metric based objective function to model the decolorization procedure.In this model,we use Log-Euclidean metric to measure the gradient of the input color image and the gradient of the transformed grayscale image.In the field of computer vision and image classification,Gaussian kernel function has efficient performance on manifold-valued data.So the Gaussian kernel is used as the penalty function in this model.A discrete searching algorithm is employed to solve the proposed model with non-negative constraint.Based on quantitative and qualitative analysis of a large number of experimental results,it is shown that our proposed model has better performance than other models.(2)Due to model based segmentation techniques too sensitive to noise and image contrast levels,this paper develops the new finite mixture model based on Dirichlet distribution and mean template for image segmentation.It has been demonstrated that a finite mixture model(FMM)with Gaussian distribution is a powerful tool in modeling probability density function of image data,with wide applications in image segmentation.Since spatial information is not considered,Gaussian mixture model is particularly sensitive to noise,light change and rain.To solve this problem,we propose a simple and efficient method to enhance its robustness.It is natural to make an assumption that the label of an image pixel is influenced by that of its neighboring pixels.We use mean template to represent local spatial constraints.Our algorithm is better than other mixture models based on Markov random fields(MRF)as our method avoids inferring the posterior field distribution and choosing the temperature parameter.We use the expectation maximization(EM)algorithm to optimize all the model parameters.Besides,the proposed algorithm is fully free of empirically adjusted hyperparameters.The idea used in our method can also be adopted to other mixture models.Several experiments on synthetic and real-world images have been conducted to demonstrate the efficiency of our algorithm.(3)Although model based segmentation techniques improve the anti-noise ability,it will lose the detail structure of the image.In order to solve this problem,this paper proposes the new fuzzy c-means algorithm(FCM)based on geodesic active contour model and Student's t-distribution for image segmentation.We present a new fuzzy c-means algorithm for image segmentation by introducing a novel spatially constrained Student's t-distribution and a new regularization term.Firstly,considering that conventional distribution models lack spatial information and the multivariate Student's t-distribution is heavily tailed,we propose a new way to incorporate spatial information between neighboring pixels into the Student's t-distribution based on Markov random field(MRF)in order to enhance robustness.Secondly,the new regularization term,inspired by the geodesic active contour(GAC)with a strong ability in capturing boundary,can preserve the details of edges and further enhance its robustness to noise and outliers by capitalizing on the local context information and edge information.Finally,in comparison to other Markov random fields that are complex and computationally expensive,the parameters are easily optimized with the EM algorithm in our proposed method.The proposed algorithm demonstrates the robustness and effectiveness,compared with other state-of-the-art method.(4)Based on the noise and detail structure in the nuclear magnetic resonance image,this paper presents the spatial constraint Student's-t mixture model based on non-local mean template for image segmentation.The mixture model is a classical and widely used tool for image segmentation.Due to the existence of noise and outliers in the images,many mixture model-based methods take the spatial information of the neighboring pixels into account to overcome the segmentation sensitivity of mixture model.Although these algorithms get certain robustness,they suffer from the over-smoothness and limited accuracy for image details.In order to further improve the robustness while preserving more edges details,we present an improved spatially varying mixture model with Student's-t distribution by introducing non-local means template for image segmentation.The proposed model is based on the Student's-t distribution,which is heavier tails and more robust than Gaussian.This template is an anisotropic weight function,defined by the space distance of all neighboring pixels and the similarity between a patch centered around it and the patches centered around the other neighboring pixels.We use this template to smooth the posterior probability map to account for the spatial dependencies among neighboring pixels.Based on Markov random field(MRF),this template is applied to the mixture model.Our proposed method directly applies the EM algorithm to optimize the parameters.Finally,experimental results obtained by employing the proposed method on many synthetic and nuclear magnetic resonance image demonstrate its robustness,accuracy and effectiveness.