Kalman filtering is an approach using the current state of the system for the prediction of its future behavior. It is an effective technique for object tracking in machine vision [12] and [13]. The potential defect regions can be true defects or false alarms induced by noises or uneven backgrounds. In the image sequence, the appearance of true defects will form a continuous trajectory, while false alarms turn out to appear randomly in irregular positions. The Kalman filtering approach can help track the true defects. The state vector and its covariance matrix are predicted using the current state (of the kth frame) and the transition matrix. The observation vector is measured by extracting the potential defect blobs of the (k+1)th frame. It is compared with the prediction in order to correct the state vector and the covariance matrix. Thus the state of system and the system covariance estimation are updated and ready for the prediction of the next frame. The procedure of Kalman filtering is illustrated in Fig. 1.