Second Order Tensor Voting in 3D and Mean Shift Method for Image Segmentation

In this paper, we present an unsupervised color image segmentation method using voting-based feature analysis and adaptive mean shift. This algorithm is based on the tensor voting approach - a unified computational framework for the inference of multiple salient structures. An unsupervised segmentation algorithm using the adaptive mean shift clustering method is applied to the reduced feature space to detect the number of clusters. A simple Euclidean distance classification scheme is used to group the pixels into corresponding color regions. Experiments are performed on color images with different complexity, and the proposed method gives satisfactory results in terms of the number of regions and region shapes.