Image coding using robust quantization for noisy digital transmission

Abstract

A robust quantizer is developed for encoding memoryless sources and transmission over the binary symmetric channel (BSC). The system combines channel optimized scalar quantization (COSQ) with all-pass filtering, the latter performed using a binary phase-scrambling/descrambling method. Applied to a broad class of sources, the robust quantizer achieves the same performance as the Gaussian COSQ for the memoryless Gaussian source. This quantizer is used in image coding for transmission over a BSC. The peak signal-to-noise ratio (PSNR) performance degrades gracefully as the channel bit error rate increases.

Approximating vector quantisation by transformation and scalar quantisation

“…Although Gaussian source is considered as the "worst case" source for data compression, which is instructive to construct a robust quantizer [16], the quantizers for Gaussian distributions are far from the optimal quantizers for other distributions. They did not consider the elliptical distributions neither, whose optimal quantizers are obviously different from those for circular distributions.…”

Robust remote sensing still image coding with low memory requirements

“…It can be shown that for Generalized Gaussian sources with shape parameter less that 2.0, such as LT coefficients, the scrambler re-shpaping in fact increases the shape parameter of the source in such a way that the new histogram is a near Gaussian one. As a result, the quantizer performance increases since the larger the shape-parameter, the better a scalar quantizer will perform [3]. Scrambling the sequences also brings a dramatic perceptual advantage since the impulsive noise is virtually eliminated.…”

“…In order to cope with the occasional errors introduced by the channel we have used the COSQ preceded by a phase scrambler (all-pass filtering) akin to that of [3]. In it, for each sequence of coefficients, a fast Fourier transform (FFT) is calculated and separated into magnitude and phase.…”

“…In [3] the impulsive noise problem is circumvented by a nonlinear all-pass filter preceding the COSQ. The all-pass filtering improves the quantizer performance and also spreads the impulsive artifacts producing images with better visual quality.…”

“…The small blocks into which the image is subdivided are transformed and the resulting transform coefficients are scrambled by means of an all-pass nonlinear filter, the output of which is quantized with a bank of channel-optimized scalar quantizers (one for each coefficient), similarly to the procedure presented in [3]. The sequence at the output of the quantizers are then encoded with rates as determined by a fast and efficient steepest descent bit allocation algorithm.…”