Markov Random Field for Road Extraction Applications in Remote Sensing Images

Bayesian methods coupled with Markovian frameworks has several applications in remote sensing images processing, such as the pixel level applications like filtering, segmentation and classification, and the higher level applications like object recognition and organization etc. This article illustrates the powerfulness of Markovian model at two levels for the road extraction problem in remote sensing images. In order to obtain the final road network, one of the low level applications is using Gaussian Markovian model to segment road target in images and then treat with the original segmentation by the line segment match method and mathematical morphology. For the sake of renewing the complete road network, one of the high level applications detects the basic road sections by homogenous texture and line segment match method, and then organizes the basic road sections in combine with context through adopting Markovian model. Experimental results show that Markovian model has good road segmentation results and high ability to interpret road network.