Hierarchical Markovian segmentation of multispectral images for the reconstruction of water depth maps

This paper presents an unsupervised method to segment multispectral images, involving a correlated non-Gaussian noise. The efficiency of the Markovian quadtree-based method we propose will be illustrated on a satellite image segmentation task with multispectral observations, in order to update nautical charts. The proposed method relies on a hierarchical Markovian modeling and includes the estimation of all involved parameters. The parameters of the prior model are automatically calibrated while the estimation of the noise parameters is solved by identifying generalized distribution mixtures [P. Rostaing, J.-N. Provost, C. Collet, Proc. International Workshop EMMCVPR’99: Energy Minimisation Methods in Computer Vision and Pattern Recognition, Springer Verlag, New York, 1999, p. 141], by means of an iterative conditional estimation (ICE) procedure. Generalized Gaussian (GG) distributions are considered to model various intensity distributions of the multispectral images. They are indeed well suited to a large variety of correlated multispectral data. Our segmentation method is applied to Satellite Pour l’Observation de la Terre (SPOT) remote multispectral images. Within each segmented region, a bathymetric inversion model is then estimated to recover the water depth map. Experiments on different real images have demonstrated the efficiency of the whole process and the accuracy of the obtained results has been assessed using ground truth data. The designed segmentation method can be extended to images for which it is required to segment a region of interest using an unsupervised approach.