Block diagram of the filtering is shown in Fig. 1. For each position of the cubic window, the DCT transform of the signal volume within the spatial-temporal cube is recursively computed from that of the previous position of the window. The signal spectra coefficients are then non-linearly modified. The inverse transform need not be computed for all pixels within the cube, since only the central sample of the cube has to be determined in order to form the output signal.
Fig. 1. Sliding cube 3D transform domain filtering
For testing the method, two sets of artificial test movie were generated. The movies were generated with two noise levels, while the filtering was done using empirical Wiener filter in 2-D domain in a 5x5 window and in 3-D domain 5x5x5 cube.
· 8-bit sequences of 4-bar targets with different contrasts on 128 background:
· 8-bit sequences of text fragment (font size 6 – font width 1 pixel – intensity level = 100) on a 180 background
Showing the applicability of the method presented two real-life thermal sequences are presented: