Robust Energy-Based Image Segmentation Using Nonparametric Joint Shape and Feature Priors

Image segmentation is a key task in computer vision that identifies important structures in images. Traditional methods like thresholding and region-based segmentation often struggle with noise and changes in shape. This paper introduces a segmentation approach that uses nonparametric joint shape and feature priors to enhance accuracy. Shape priors are learned from the MNIST dataset, and feature representations are extracted through Principal Component Analysis (PCA). We formulate the segmentation as an energy minimization problem that combines data fidelity, shape, and feature terms. Experimental results, measured using Dice coefficient and Hausdorff distance, show the effectiveness of this method.