Underwater target identification using GVF snake and zernike moments

This paper is focused on the development of robust object segmentation and shape-dependent feature extraction methods for automatic water target classification and identification using electro-optical imagery data. The sensor used for acquiring the data is the Streak Tube Imaging Lidar (STIL) that offers both range and contrast images with high resolution. In this paper, the gradient vector flow (GVF) snake is employed to segment the detected objects. The snake converges to the actual object boundary and provides a closed contour of the object even when some of the edges are missing. To reduce the distortion as a result of missing edges, the union of the binary silhouettes for contrast and the range images is obtained. Zernike moments are then computed for the combined silhouette of the segmented object. These moments provide shape-dependent features with high discriminatory ability, which are invariant to object rotation, translation and size scaling in the image. This set of features is then used for target identification from the STIL imagery data. To aid discrimination of different objects with potentially similar shape dependent features, mean and variance of the contrast and range images are also computed within the closed contour and then used as additional features for classification. Then the extracted features are applied to a multi-layer back-propagation neural network (BPNN) that performs target classification/identification. Different neural network structures are tried to determine the optimum classifier. The effectiveness of the developed algorithms is demonstrated on several data sets and the corresponding confusion matrices are also developed.