Conditional Random Fields for Image Labeling

Figure 2

Some examples of labeling problems in computer vision. For stereo matching, the goal is to find the corresponding pixel in one image given a pixel in another image. Its label set is the differences (disparities) between corresponding pixels. For image segmentation, its goal is to partition an image into multiple disjoint regions with region IDs as its label set. For image restoration, it tries to “compensate for” or “undo” defects which degrade an image, and its label set is restored intensities or color.

(a) Stereo matching

(b) Image segmentation

(c) Image restoration