Self-Training Statistic Snake for Image Segmentation and Tracking

In this work we propose a new supervised deformable model that generalizes the classical contour-based snake. This model is defined to deform in a feature space generated by a set of Gaussian derivative filter responses. The snake selects and classifies image features by a parametric vector that gives the direction in the feature space minimizing the dissimilarity between the learned and found image features and maximizing the distance between different contour configurations. Each snake curve patch is devoted to search for a special contour configuration. The classes coresponding to different contour configurations are obtained by means of a statistic supervised learning technique using samples of different contour and no contour points. The snake starts with a large set of Gaussian filters that is reduced by means of principal component analysis in a supervised way to optimize it in the feature search.

• Type: Conference
• Year: 1999
• Where: ICIAP
• Authors: Xose Manuel Pardo, Petia Radeva, Juan José Villanueva
• Keywords: Contour Configurations, Feature Space, ICIAP 1999, Image Processing, Supervised Deformable Model