Mine detection in sonar images with minimal user's input

A parameter-free and a priori-information-free preprocessing of sonar images is proposed, which permits a ranking of local extrema in the image according to their likelihood to be a mine-like object. It is shown that an acceptable fully automatic detection algorithm can be built on a variational method which estimates shape information of the possible mines. This algorithm does not use any a priori information on the type of mine or range distance or background type and works without any change on both sonar databases we had available (Sonar0 and Sonar3). It therefore can be used as a detection algorithm without any information request the user or designer. Its results could be fed into a classification algorithm like the one proposed in Ref. 1. We also think that the features computed by this variational method could serve for both the detection step and the classification step, thus reducing the number of designer's parameters and opening the way to a parameter-free detection-classification algorithm.