Markov Random Field Labeling of InfraRed Thermal Images: Applications in Industry and Veterinary Medicine

In this paper, we propose an efficient approach for object segmentation in IR thermal images. Markov random field (MRF) is used for efficient segmentation that incorporates spatial information through priori information of the local structure present in the IR image. MRF is a conditional probability model that uses the statistical correlation of pixels among its neighborhood. The thermal parameters associated to each label in the IR image are derived based on K-means, unsupervised learning algorithm as initial label. Under the MRF segmentation framework, an energy function is formulated that comprises of a data driven term and a regularizing term involving the prior knowledge of the label associated. Upon minimization of the energy function results in the accurate labeling of different classes. To show the efficacy, the proposed approach is applied in the field of veterinary medicine to detect and segment the foot-and-mouth disease (FMD) of infected cattle. In another application, a non-invasive corrosion monitoring and assessment is demonstrated with the MRF labeling of IR images.