Method for visual optimisation of embedded block codes to exploit visual maskingphenomena

摘要

An improved method is described for lossy compression of images, so that visual distortion is reduced for a given compressed bit-rate or, equivalently, a lower bit-rate is required for a given level of visual distortion. The image is decomposed using a Wavelet transform of other space-frequency transform. The frequency bands are partitioned into small blocks (e.g. blocks 32 by 32 samples). The blocks are independently quantized and coded using an embedded block coder, so that each block bit-stream contains a large number of finely spaced truncation points. A visual distortion measure is computed for each block at each truncation point, where the metric is sensitive to the masking properties of the Human Visual System; that is, quantization errors in regions where the relevant frequency band already contains substantial activity are assigned a smaller visual distortion than comparable quantization errors in regions where the frequency band exhibits little activity.