一种模拟视觉机制的图像分割模型

A Visual Mechanism Inspired Model for Image Segmentation

提出一个小波域多尺度马尔柯夫随机场模型用于模拟视觉系统在图像分割中的若干功能。针对人类视觉系统具有特征检测器、等级层次性、双向连接性、学习机制等功能,对输入场景,该模型用小波变换提供该场景图像的稀疏表示,模拟特征检测器功能;用金字塔结构模拟等级层次性;用两类信息流模拟双向连接性,分别刻画自底向上的输入图像特征提取过程以及自顶向下的反馈过程;用迭代过程模拟学习机制;采用多尺度马尔柯夫随机场模型实现图像分割。实验表明,该模型对真实采集到的不同类型的生物医学图像进行分割,取得优于一些传统分割算法的结果。

In this paper,a multiscale Markov random field（MRF） model in the wavelet domain was proposed by simulating several image segmentation functions of the visual system.Human visual system（HVS） has feature detection ability,hierarchy,bidirectional connection,and self-learning mechanisms.For an input scene,our model provided its sparse representations using wavelet transforms（WT） to mimic the feature detection ability,and used pyramid framework to mimic hierarchy.In the framework of the model,there were two information flows that were used to mimic bidirectional connection,i.e.,a bottom-up procedure to extract input features and a top-down procedure to provide feedback controls.In addition,the iteration in procession was the simulation of self-learning mechanisms,and the multiscale MRF was the tool for image segmentation.The quality of the framework was tested and compared to some classic image segmentation algorithms.Results showed that the proposed model obtained improved data than those obtained by classic image segmentation algorithms.