The method for segmentation of hand from complex backgrounds is presented in [17]. The scheme first utilized motion information to detect the rough position of the hand. Then, we applied the prediction-and-verification scheme based on the attention images of the multiple fixations to segment the hand from the complex backgrounds. We tested the system with 805 images (161 sequences) which were not used in the training. The results are summarized in Table 1. The time was obtained on a SGI INDIGO 2 (MIPS 4400 and clock 150 MHz) workstation. Fig. 12 shows some segmentation results.
Table 1: Summary of the segmentation results
Figure 12:
The results of the segmentation are shown after masking off the background.