Applications of hidden Markov chains in image analysis

Oslo in 1988. She is currently working as a researcher at the Norwegian Computing Center and has been involved in several projects concerning document image analysis and machine vision. Her research interests lie mainly within the various applications of statistical pattern recognition. Summary In this paper, we demonstrate that hidden Markov chains have a potential for use in diierent image analysis problems. Four examples are presented where the data source varies from a 1-band image to a live video sequence, and the applications range from machine vision to video surveillance. The rst example is fetched from text recognition. Here we use hidden Markov chain models for each character to obtain simultaneous segmentation and recognition of printed words from a grey level image. The second example is a machine vision problem, where the issue is to recognize return bottles in crates imaged from above. A laser range scanner is used to aquire the images and the hidden Markov chain models are used to describe scanlines through the bottles. In the remaining two examples, we have a sequence of images. In the rst of these examples, the problem is to classify tumors as malignant or benign based on a dynamic Magnetic Resonance Image (MRI) sequence. The hidden Markov chain model is here used to describe the evolution of the image intensity with time after injection of contrast agent in the case of a malignant tumor. The last example is a live video sequence, where the video images are used for traac surveillance. The problem is to detect (and count) vehicles passing the surveillance site. Two detectors are used for this task, and the course of events reported when a vehicle is passing gives a sequence of two-dimensional feature vectors that is modelled as a hidden Markov chain. Even though images are 2-dimensional, we have shown that the problem which is to be solved often is of a 1-dimensional character. Hence, hidden Markov chains may be applicable and their several strong points also can be exploited in image analysis. Abstract In image analysis, 2-dimensional Markov models, i.e. Markov eld models, have been applied for segmentation purposes, but except for the area of text recognition, the application of hidden Markov chains has been rare. Through four very diierent examples, this paper demonstrates the applicability also for hidden Markov chains in image analysis, and shows that the problems of image analysis often may have 1-dimensional …