Machine Vision Design And Application Based On Markov Random Field

The Markov random field is one of the important models in machine vision and digital image processing.The theoretical research on Markov Random Field is also very active.The theory has always had new problems emerging.Its theory is in line with people's intuitive understanding of natural things.In the field of digital image processing,the Markov Random Field itself does not possess a certain ability,nor is it a method in image processing.It describes a collection of certain things that have Markov nature,can make good use of prior knowledge of things,and is embedded as a prior model into some algorithms that have specific effects.Because of its model parameters are few and can describe the spatial information and have a set of perfect mathematical theory system and other advantages have been concerned.Digital image processing mainly has two purposes: one is to obtain or generate images that are suitable for people to recognize and watch;the other is to help computers understand and identify the goals in images.It divides the image into areas of each characteristic,and can extract the objects of interest,laying the foundation for the subsequent steps of characterizing and estimating measurements.Markov Random Field provides a universal framework.In image processing problems,it can easily use a variety of different features,and can also fully consider the interaction between different types of pixels.The energy function of Markov random field considers the dependency relationship between adjacent pixels,establishes the correlation between pixels and their neighbors,and increases their probability of falling into the same area.Successfully translated the image segmentation problem into a predictable and controllable problem.This paper mainly studies the Markov image segmentation algorithm based on conditional iterative algorithm(ICM).This algorithm uses random pre-classification to segment the image,has many iterations,and easily causes some regions to fall into the local optimal solution.At the same time,the accuracy of the segmentation of the edge contour and other details is insufficient.However,its theoretical basis is simple and the model is easy to implement.The coupling coefficient can be changed,the number of classifications can be defined,and the number of iterations.In order to achieve a better segmentation result,the adjustment of different scene parameters is more flexible and has better plasticity and utilization value.This paper combines the Markov random field image segmentation algorithm and the graph-based image segmentation(GBIS)algorithm.The GBIS algorithm considers the YCbCr distance of pixels,and quickly divides the image by a simple greedy clustering algorithm,which causes uneven illumination and the segmentation of the color gradient region has better results,but the division of the edges is more ambiguous.In this paper,the algorithm is combined with the traditional Markov image segmentation algorithm.The advantages of the two algorithms are successfully used,the number of iterations is greatly reduced,the result is faster and stable,and the original Markov image segmentation algorithm method is improved to achieve local in the best case,the improved algorithm overcomes the effects of illumination and difficult-to-recognize problems in the color gradient region,and has better robustness.The comparison of the results of a large number of simulation experiments on different scenes such as avatars,buildings,and fields shows that the proposed algorithm is better.