CV Image Segmentation Model Combining with Local and Global Features of the Target

EasyChair Preprint 3646

Abstract

With the development of remote sensing satellite technology, high-resolution remote sensing images continue to emerge.Automatic target extraction from remote sensing images with more information and complex background has become an urgent problem to be solved.The traditional image segmentation method mainly depends on the image spectrum, texture and other underlying features, in image segmentation tasks, shadow, occlusion and various interference in the image increase the difficulty of segmentation, and lead to unsatisfied results.For this reason, according to the specific target type, a CV (ChanVest) image segmentation model combining with local and global features of the target is proposed.Firstly, the deep learning generation model-CRBM(Convolution Restricted Boltzmann Machine), is used to represent the global shape features of the target and to reconstruct the shape of the target.Secondly, the edge information of the target is extracted by Canny operator, and a new shape  constraint term which combines the local edge and global shape information is obtained by symbolic distance transformation  is prop0sed.Finally, the CV model is used as the image target segmentation model, and new constraints are added to obtain the CV (ChanVest) remote sensing image segmentation model which combines the local and global features of the target.The experimental results on remote sensing dataset Levir-oil drum, Levir-ship and Levir-airplane show that the proposed model can not only overcome the noise sensitivity of CV model, but also segment the target completely and accurately in the case of limited training data, small target size, occlusion and complex background.

Keyphrases: CV model, Convolutional restricted Boltzmann machine, Shape prior, deep learning, image segmentation

@booklet{EasyChair:3646,
  author    = {Xiaohui Li and Xili Wang},
  title     = {CV Image Segmentation Model Combining with Local and Global Features of the Target},
  howpublished = {EasyChair Preprint 3646},
  year      = {EasyChair, 2020}}