Texture analysis of sidescan sonar data

Describes the use of texture analysis techniques for extracting information from sidescan sonar images of the seafloor. High resolution sidescan images are often highly textured due to the presence of sedimented regions of seafloor (sand ripples, mud, gravel, etc.). Two techniques are presented for the classification of these images into regions of different sediment type. The first of these is based upon the fractal model, however it is demonstrated that calculation of the fractal dimension is not necessary and that consideration of the directionality of the textures is important. The second technique is based upon the theory of spatial processes, and is concerned with the distributions of grey-level values in the textures. Results for both techniques are presented for samples of sidescan data. Another application of texture analysis is presented for the detection of man-made objects in sidescan images of the seafloor. These objects are regarded as being disturbances in the background texture and may be detected by computing probabilities from the texture model. Results are presented for samples of real sidescan data. Finally, a novel technique for compressing sidescan data is presented. This technique uses the information obtained from the sediment classification and object detection schemes to generate an image containing synthetic textures following the original sediment boundaries, with the targets of interest included at the appropriate positions