The order of PARCOR images

The order of PARCOR images is important factor of recognition ability. By increasing the order of PARCOR images, information extracted from the sequence of the images also increases. To determine the necessary order of PARCOR images for gesture recognition, we have performed experiments using HLAC features extracted from different orders of PARCOR images. In this experiments, PARCOR images of $16 \times 12$pixels were used. The number of samples is 450 (50 samples $\times$9 categories). The recognition rates were estimated by leave one out method. .

The results are shown in Table 1. By increasing the order of PARCOR images, the estimated recognition rates also increased. But HLAC features extracted from the first and second order PARCOR images seems sufficient for recognition of these simple gestures. By adding HLAC features extracted from the average image, the recognition rate was further improved.

Since enough recognition rates was obtained using HLAC features form small number of PARCOR images, we can obtain HLAC features of PARCOR images in video rate by using online estimation method of the autocorrelations.

 
 

Table 1: Recognition rates with different orders.

Orders of PARCOR images	Rate (%)
1st	86.4
1st and 2nd	91.1
1st and 2nd + average image	98.9