中文摘要：

图像分割问题是计算机视觉领域研究的基础性问题。针对实际图像中无纹理对象的浅阴影分割过程,通常会假设这些对象为同性质分段状态,而基于这种假设条件的图像分割方法有可能会产生图像分割偏差。本文方法通过放宽同性质均匀假设条件,针对图像强度进行不均匀平滑处理。本文所提算法应用图像中对象强度的分布一致性,采用新型平滑度计算方法来提高图像分割效果。根据待分割图像局部结构来计算分布一致性,图像分割过程中则应用贝叶斯框架。同已有研究成果比较了Hessian矩阵和方向张量的分割效果,通过在人工图像和真实图像上的实验结果表明,本文所提算法相较全局阈值与多层次逻辑马尔科夫随机域模型能够得到更好的图像分割效果。

英文摘要：

Image segmentation is a basic problem in computer vision research.For the light shadow segmentation process of no texture objects in actual images,we usually assume that these objects are in segmentation with same nature.However,it will generate segmentation deviation basing on this assumption condition.In this paper,our method can be used for image intensity non-uniform smoothing through broadening same nature uniform assumption condition.In order to improve the effect of image segmentation,the proposed algorithm uses uniform distribution of object intensity in the image and adopts a new smoothness calculation method.The Bayesian framework is applied in the process of image segmentation.We calculate the uniform distribution according to the local structure of image segmentation.Based on artificial and real images,the paper had been conducted comparative evaluation of Hessian matrix and orientation tensor segmentation.The experimental results show that our proposed algorithm significantly outperforms the global threshold and Multi-level logic Markov random field model.