Texture segmentation based on the use of the structure tensor and the wasserstein distance

xiexiaozhen; wu jitao

doi:10.11834/jig.20100911

Views : 0 下载量: 282 CSCD: 0

PDF
Export
Share
Collection
Album

Texture segmentation based on the use of the structure tensor and the wasserstein distance
Vol. 15, Issue 9, Pages: 1345(2010)
Published Online：27 September 2010，

Published：2010
DOI： 10.11834/jig.20100911
稿件说明：

移动端阅览

DOI：

xiexiaozhen, wu jitao. Texture segmentation based on the use of the structure tensor and the wasserstein distance[J]. Journal of Image and Graphics, 2010, 15(9): 1345. DOI： 10.11834/jig.20100911.

摘要

基于区域的非参数活动轮廓模型被广泛地应用于图像分割领域。在结构张量和图像灰度直方图基础上建立的新模型就属于这个范畴

它被用于灰度和彩色纹理图像的分割。首先，对经典的结构张量进行扩展，使其能表达原始纹理图像的梯度信息和色彩信息。利用扩展的结构张量，使一幅图像生成多个特征通道，从而提取更多的纹理特征。然后，在各个通道中，利用像素点的局部灰度直方图的Wasserstein距离信息分割图像，使图像不同分割区域的灰度分布近似同态。最后，为了有效地降低模型的数值计算复杂度，一些数值计算技巧被引入模型的计算过程，产生了很好的效果。通过大量的实验对比，利用新模型能得到更精确、更稳定的分割结果。

Abstract

Nonparametric region-based active contour models have been widely used in the field of image segmentation. The proposed new model which is based on the use of the structure tensor and the Wasserstein distance belongs to this category. First

the classical tensor structure method is extended in order to express the gradient and color information of the original images. Using the extended tensor structure method

an image can yield more feature channels and more features of the texture can be extracted. Then

we use the Wasserstein distance between each pixels’ local distribution functions of the image intensities to segment the image domain so that the local histograms within each region are approximately homogeneous. Some numerical mechanisms are introduced into the approach

which obviously reduce the computational complexity of the model. Comparing with related studies

experimental results show that proposed model is more precise and stable.