Experimental study on the discrete trigonometric transform (DTT) wavelet-like decomposition stage with windowed filters

摘要

In this paper an experiment designed and performed to verify the possibility of efficient windowing utilized in time domain with the Mallat type wavelet decomposition scheme is described. The main idea is that the used filters are designed not in the Fourier domain but in a domain of any of the 16 DTTs. It is demonstrated that the windowing technique, which reduces the number of the required computations up to 34 times, results in the error with statistically assessed PSNR of ca. 40 dB (from 36 to over 50 dB). The PSNR is obtained from MSE computed with an assumption that the processed signal is a realization of the first order Markov process with a given intersymbol correlation. If necessary, the reconstruction is assumed to be perfect.