Research And Application Of Image Region Segmentation Algorithm

Image segmentation is the most important step in the process of digital image processing,which plays a key role in computer vision.Because there is no unified theory for guidance,although there are thousands of methods of image segmentation,but there is no way to deal all the images.Traditional classical algorithm has edge detection method,region segmentation method and threshold segmentation method,because of their limitations and lack of prior knowledge reference,it is difficult to improve the accuracy of image segmentation.Therefore,it is of great significance to study the image region segmentation deeply.This paper mainly studies the region segmentation algorithm of image and improves the threshold segmentation,level set method and watershed transform.The improved method is applied to segment gray image,medical image and remote sensing forest image.The research contents are divided into the following three parts.For the recursive binarization segmentation,the basic idea of this algorithm is to select the whole image threshold,which will result in the loss of the detail information of the image.For the image with uniform gray value,the method can achieve better segmentation effect,but the segmentation result will appear pseudo contour for the image with the large differences between the target region and the background region.In order to avoid the loss of image detail information and solve the pseudo contour phenomenon,the recursive binarization segmentation is improved.The whole image is modularized,the initial threshold and estimate threshold are selected in each module,and the image is segmented according to the set iterative condition.In the process of image processing,the improved method considers the global and detail information of the image and avoids the loss of information and the pseudo contour effectively.The level set method of M-S model depends on the global information of the image homogeneous region to segment the image,thus it is low time efficiency in segmentation process.To improve the computational efficiency,this method has improved by many researchers in the field of image processing.Based on the C-V model,the paper discusses three kinds of improved segmentation evolutionary algorithms:the algorithm based on the C-V model without the regularization term;the algorithm of replacing the Dirac function ?(?)by |??| for the purpose of better global optimization;the algorithm of adding the local gradient term suitable for dealing with image of weak edges and edges fracture.Finally,three examples is presented to further illustrate the efficiency and the range of applicability of the algorithms.In the paper,an improved homomorphic filter is introduced to extract the single tree crown contour of forest remote sensing images in dense forest and sparse forest areas.Based on the watershed algorithm,an improved mark control watershed segmentation method is presented to extract the tree contour.The improved homomorphic filtering can remove the mixed noise and does not destroy the edge information.The preprocessed image edge information is more complete.Combining with the results of artificial recognition,the improved method is compared with the traditional method of segmentation.The improved method with high accuracy has obvious improvement on the crown contour of the sparse forest area and is easy to implement.