Multispectral-image compression based on tree-structured Markov random field segmentation and transform coding

This paper presents a new compression technique for multispectral images. The proposed encoding algorithm is based on two steps: segmentation and transform coding. The segmentation step is based on a hierarchical tree-structured Markov random field model for the image, which is able to effectively take into account the spatial dependencies. After segmentation, class-adapted transform coding is used to decorrelate information both in the spectral and spatial domain. Simulation results show that the proposed technique exhibits a significant performance gain at very low bit rates, while assuring a satisfactory image quality.