#159: Multiscale focus driven segmentation using elastic body splines

S. Meena, R. Pelapur, V. B. S. Prasath, and K. Palaniappan

Proc. SPIE Conf. Geospatial Informatics, Fusion, and Motion Video Analytics VI, 2016

interactive, segmentation, focus

PlainText, Bibtex, Google Scholar


We study a method for extracting blurred or sharp regions of interest (ROI) that could help in initializing an interactive segmentation method using elastic body splines. Accurate salient region detection and segmentation can help many vision applications. In order to detect the ROI we first classify each pixel in an image as either sharp or blurred based the absolute magnitude of the Frobenius norm of eigenvalues from multiscale Hessian matrix. The Frobenius norm would be low in areas of blur relative to the areas that are sharp and in-focus. We use this basic property to mask out the ROI and initialize our novel elastic body splines (EBS) based interactive segmentation technique. Elastic body splines belong to a family of splines and have been applied for the task of biomedical image registration. It models the elastic deformation of homogeneous isotropic elastic body subjected to external forces. We utilize a logistic regression fusion strategy for combining both focus based saliency map and EBS segmentation. Our initial set of experiments shows promise and has improved the quality of the segmentations over the Hessian based ROI detection and EBS results on natural images.