Classification of side-scan sonar images through parametric modeling

Techniques for the classification of side-scan sonar images in general must rely solely on texture analysis due to the lack of multispectral information. The authors have investigated the use of parametric texture modeling for side-scan image classification. Autoregressive (AR) and autoregressive, moving-average (ARMA) models are applied to the image, The model parameters are estimated adaptively on a local basis and then used as input to a standard maximum-likelihood classifier. Examples are provided and the results are compared to those obtained through a previously proposed technique based on sub-band analysis.< >