Error detection and correction of vector quantised digital images

The performance of a low bit-rate digital image codec based on a product vector quantiser (VQ) is examined in a noisy transmission channel environment. The VQ codec is a separating gain/shape type providing good data compression at about 1.25 bits per pixel. For small image subblock sizes, there is sufficient interblock correlation in the mean, standard deviation and shape information to facilitate crude error detection mechanisms to be implemented. We reinforce these error detection capabilities with a subblock boundary continuity testing algorithm. Erroneous subblocks are then estimated by extrapolating intensity contours from neighbouring subblocks. Using such an error detection and correction technique, noisy recovered VQ images can be considerably improved. At a bit error rate of 0.01, we obtained an improvement in the normalised mean square error of about 0.013, or equivalently about 6 dB in signal-to-noise ratio.