基于 MRF 与纹理特征的图像深度信息估计

摘要

提出了一种基于马尔科夫随机场（ MRF）模型与多尺度纹理特征的单幅图像深度信息估计方法，该方法采用了Laws滤波器分别对图像的边缘、梯度、点进行滤波，捕捉二维场景图像中不同尺度的纹理能量以获得深度信息的特征．并根据纹理特征在不同尺度范围的不同值，计算出纹理线索与场景深度间的概率关系，在此基础上，构建MRF概率模型．MRF模型通过分析邻域系统和设计迭代准则很好地描述了纹理特征与场景深度之间的关系，最后通过迭代算法获得二维场景图像的深度信息．实验结果表明，该方法对场景深度信息的提取具有较好的效果，对于二维场景图像的场景结构、空间布局的约束较少，算法鲁棒性好．%This paper proposed a single image depth estimation method based on Markov ran-dom field ( MRF) and multi-scale texture features, which can better describe the relation-ship between the texture features and the scene depth.Captured the texture energy of differ-ent scales as the features of the image using Laws filters to image texture gradients, edge and point, and calculated the probability that the relationship between texture clues and scene depth.Based on this, the MRF model can be established.By analyzing the neighborhood system and designing the iteration criterion, MRF model described the relationship between the texture feature and the scene depth, and got the image depth information through the iter-ating algorithm.The experimental results showed that it was effective for image depth infor-mation estimation.Meanwhile it had good robustness because it has less limitation to the scene structure of the image.