Image Segmentation using Tensor Voting

Elaine Kang


Abstract

Image segmentation entails the division or separation of the image into regions of similar properties or homogeneous regions. The commonly used properties for image segmentation are intensity, color and texture. Homogeneous region extraction is essential for further image analysis, understanding and interpretation such as object recognition and moving object tracking. Most of the segmentation methods use a single property to define homogeneity and extract regions, while humans perceive regions by the simultaneous integration of several visual cues. In this talk, we present a tensor voting based image segmentation method that integrates several properties naturally in an N-D tensor framework and begets substantial information for region clustering, and we show promising preliminary results.


Maintained by Philippos Mordohai