Research On Moving Object Segmentation Method In Complex Background

In the field of image processing, the segmentation and tracking of moving object in video sequences is a hot research topic in recent years,various of segmentation methods can be used according to different situations of moving objects and its background.In this paper,the author use Markov Random Field Model as main algorithm of segmentation and tracking of moving object in a stationary complex background,but it has its weaknesses in high time complexity, slow processing speed and inaccurate estimation for parameters, these weaknesses have always been the biggest obstacles which hinder the method form applying in project practice. In addition,traditional MRF work well in simplified background,but it can't achieve the ideal effect in complex background,such as sudden change of background ray, change of background objects (containing trees blowing in the wind and fountains).Therefore, in this paper improvement was made focused on those weaknesses.Most video image processing questions come down to the questions of obtaining labels.In another word, the solution of video image processing question is to distribute labels to every pixel in image. Moving objects segmentation come down to optimization of initial labels in the observations and a series of constrained conditions.It is intended to achieve optimized initial labels.Therefore, the results of initial labels have important influence on MRF segmentation algorithm. If it was different obviously between initial labels achieved and actual results, the perfect result can't obtained,and it goes against algorithm converges in global optimum.In order to get better accurate initial labels,the paper make improvement on Gibbs energy function.It obtained the original initial labels by R, G, B color vectors, then reassigned labels for the original initial labels by improved energy function and obtained the final initial labels. Namely, it was one time optimization treatment for the original initial labels.In order to improve the efficiency of image segmentation based on MRF,the paper starting from removing the redundant computational complexity. Firstly ,the symmetric difference algorithm was used for determining a small region including moving object, and then , MRF segmentation algorithm was used for determining the exact boundary of moving objects in the small region. In this way,it can decreased the number of pixels taked part in computing to under the half of the former,it not only increased computing speed,but also removed noise interference beyond the small region.The experiment showed that this improved algorithm can get ideal segmentation result and good real-time performance...