Real-time modeling of image sequences based on hidden Markov meshrandom field models

摘要

The image modeling problem is discussed under the assumption thatimages can be represented by third-order. hidden Markov mesh randomfield models. The modeling applications comprise restoration of binaryimages, compression of image data, and segmentation of gray-level imagesand image sequences under the short-range motion hypothesis. Coherentapproaches to the problems of image modeling and estimation of modelparameters are outlined. A labeling algorithm based on a maximummarginal a posteriori probability criterion is proposed. Criticalaspects of the computer simulation of a real-time implementation arediscussed in detail. A learning technique by which the model parameterscan be estimated without ground truth information is developed.Extensive experimentation with both static and dynamic images from avariety of sources is discussed