PROJECT TITLE :

Multimodal Change Detection in Remote Sensing Images Using an Unsupervised Pixel Pairwise-Based Markov Random Field Model

ABSTRACT:

The multimodal change detection (CD) problem in remote sensing imaging is addressed using a Bayesian statistical approach. Furthermore, the multimodal CD problem is formulated as an unsupervised Markovian problem. This Markovian model uses a pixel pairwise modelling observation field and a pair of bitemporal heterogeneous satellite images as its primary originality. To avoid this, we can use modelling techniques that allow us to rely on a robust visual cue that is nearly invariant to the imaging (multi-) modality. We first utilise a preliminary iterative estimating technique that takes into consideration the diversity of laws in the distribution mixture and predicts the parameters of the Markovian mixture model in order to leverage this observation cue in a stochastic likelihood model. An optimization procedure based on the previously calculated parameters is used to construct the MAP solution of the change detection map, a stochastic optimization process. Experiments and comparisons with various imaging modalities show that the proposed approach is robust.