Edge-Labeling Using Dictionary-Based Relaxation

Edwin R. Hancock, Josef Kittler

Abstract: An improved application of probabilistic relaxation to edge labeling is presented. The improvement derives from the use of a representation of the edge process that is internally consistent and which utilizes a more complex description of edge structure. The application uses a dictionary to represent permitted labelings of the entire context-conveying neighborhood of each pixel. Details are given of the dictionary approach and the related representation of the edge process. A comparison with other edge-postprocessing strategies is provided.<