A texture analysis of 3D GPR images

In this thesis, image processing algorithms are applied to 3D GPR images, in order to improve the detection capabilities of a radar system. Detection based on the magnitude of the reflected signals may miss weak targets. On the other hand, an analysis of the texture properties of a target, i.e. the repeating pattern all over the surface, which is independent on the signal intensity, discriminates it better from the clutter. The texture analysis algorithm applied to 2D and 3D radar images is called Texture Feature Coding Method (TFCM). It highlights neighboring volume pixels (voxels) with high correlation and it has been applied iteratively to global and local volumes of the 3D image, in order to improve the detection of weak targets. The measurement of the correlation between neighboring voxels is based on a tolerance value, and an threshold algorithm to automatically detect this value has been customized. Image visualization is performed with automatic threshold selection, extracted from the histogram of the 3D images. The algorithm has been applied to images of landmines or mine-simulant objects laying on the surface, giving remarkable results. The method is successfully able to detect the targets and to highlight their edges, allowing a realistic visualization of the shapes of the targets. Further research in this direction is suggested: tests on buried targets should be performed in order to validate the algorithm. A degradation of the results is expected when buried targets are used, however, texture features can be extracted and object classification techniques can be used in order to discriminate between clutter and targets.