Summary: | 碩士 === 大同大學 === 電機工程研究所 === 85 ===
A low bit-rate subband coder is presented in this thesis. This design is based on removing perceptual redundancy according to basic characteristics of the human visual system (HVS) in the spatial and temporal domain. In order to quantify the perceptual redundancy inherent in the video signal, we use a perceptual model that can estimate the just-noticeable-distortion (JND) profile for a sequence of video frame, and by which to evaluate the perceptual important signal to be encoded.
This thesis presents a subband video coding scheme based on a traditional QMF analysis. Since the original image is decomposed into a number of frequency bands by the subband filters, the JND profile of the original image must be also decomposed into component JND profiles for different subbands according to the modulation transfer function (MTF) of HVS.
The lowest band, which has the most important information, must be coded in detail. But in order to reduce image bit-rate, we select the perceptual important signal to be coded with the spatio-temporal JND profile of the lowest band. In the higher bands, we utilize the HVS properties and the same selection method as the lowest band does. A computable fidelity measure is exploited to evaluate the quality of the compressed image in terms of the perceptible distortion energy. The fidelity criterion, as termed by peak signal-to-perceptible-noise ratio (PSPNR), is presented in this thesis. simulation results have a better quality as the average bit rate is below 0.35 bpp. The PSPNR of the "football", "Miss American", and "salesman" sequences at the 0.24~0.27 bpp, 0.07~0.12 bpp, and 0.32~0.38 bpp are 37~40 dB, 40~43 dB, and 34~35dB, respectively.
|