Automatic color blob segmentation and fast arbitrary shape tracking

Efficient image segmentation is the first step towards object-and content-based video applications. Tracking of segmented regions (blobs) is even more important and challenging due to abrupt object motion, nonrigid shape distortion, object and scene occlusions, and camera motion. This paper proposes a novel system for automatic and efficient color blob segmentation in image spatial domain and fast arbitrary-shape blob tracking in video sequences. Firstly, a dual-threshold Connected Component Labeling (CCL) algorithm is proposed for efficient blob extraction in grey scale images. Then an unsupervised image segmentation method is designed using the CCL and meanshift principle. With the blob information obtained from the image segmentation procedure, each pixel within a color region is then assigned a region index number which is linked with two blob parameters: the size and the average color intensity. To track an arbitrary-shape blob through a video sequence, a novel algorithm based on Region Index Map (RIM) is employed instead of using traditional motion tracking method to achieve efficient tracking. The experimental results show that the proposed system provides automatic and efficient spatial segmentation. Moreover, robust temporal tracking for fast-moving regions is demonstrated without computationally expensive motion estimation.