Adaptive Contextual Energy Parameterization for Automated Image Segmentation

Image segmentation techniques are predominately based on parameter-laden optimization processes.The segmentation objective function traditionally involves parameters (i.e. weights) that need to be tuned in order to balance the underlying competing cost terms of image data fidelity and contour regularization. In this paper, we propose a novel approach for automatic adaptive energy parameterization. In particular, our contributions are three-fold; 1) We spatially adapt fidelity and regularization weights to local image content in an autonomous manner. 2)Wemodulate the weight using a novel contextual measure of image quality based on the concept of spectral flatness. 3) We incorporate our proposed parameterization into a general segmentation framework anddemonstrate its superiority to two alternative approaches: the best possible spatially-fixed parameterization and the globally optimal spatially-varying, but noncontextual, parameters. Our segmentation results are evaluated on real and synthetic data and produce a reduction in mean segmentation error when compared to alternative approaches.