基于DRF-MAP模型的单目图像深度估计的改进算法

Depth Estimation from Monocular Images Based on DRF-MAP Model

图像深度信息获取是机器视觉领域的活跃研究课题之一。将图像深度估计问题归结为模式识别问题,以单目图像深度为模式类,在多尺度下从图像块中提取绝对和相对深度特征,并选择表征上下文关系的DRF(Discriminative Random Field)方法来表述某图像块的深度和其邻域深度之间的关系,从而构建起基于DRF-MAP(Maximum a posteriori)的单目图像深度估计模型。通过实验,得到了一类单目图像对应的深度图像,从而证明了单目图像深度估计模型对应的改进算法的有效性。

Depth Estimation is an active research area in computer vision. We viewed this problem as an issue of Pattern Recognition. The depth of monocular images is the pattern to be classified. Based on DRF-MAP Model, we combined absolute features and relative features of the monocular images in multi-scale, modeled the relationship between depths and features and those relations among depths at different patches in the image. Our algorithm is able to receive depth-maps for certain kinds of monocular images, which demonstrates the algorithm＇s effectiveness of the model.