Conditional Random Fields for Image Labeling

Figure 1

An example of image labeling. An image in (a) is a set of pixels with observed intensities for each . A labeling shown in (b) assigns some label to each pixel . Such labels can represent depth (in stereo), object index (in segmentation), original intensity (in image restoration), or other pixel properties. Thick lines in (b) show labeling discontinuities between neighboring pixels [5].

(a) An image

(b) A labeling