Image segmentation is an important field of image processing. It consists in partitioning the image into non-overlapping meaningful homogenous regions i.e. flat regions, movement (stereo, optical flow), model-based, texture, color, ... etc. This has been widely used in different applications, for instance, medical images and robot vision. This work focuses on two main themes. The first is related with image segmentation problem and the second is about an application of segmentation methods to image and video editing. In the last decade especial attention has been paid to segmentation methods that produce a measure of belonging to classes, instead of classical segmentation methods that obtains a label map. The first kind of methods is known in the literature as “soft” segmentation methods while the second group is called as “hard” segmentation methods. This work presents a general framework for “soft” segmentation with spatial coherence through a Markov Random Field prior.