Although the wavelet transform has become a popular tool for multiresolution storage and progressive transmission, considerably less attention has been paid to the impact of storage or transmission (channel) errors on the retrieved and reconstructed picture. The IP team proposes to investigate these errors and to develop algorithms which reduce their effects.
Image compression schemes often include, as a final stage of processing, a variable word length coding operation. The bitstream transmitted by the channel or stored to disk (tape), thus, consists of codewords which can only be unambiguously decoded, if they are valid codebook entries. If, however, errors occur and synchronization is lost, potentially catastrophic effects upon the reconstructed image can happen. Traditional forward error correcting methods offer one solution to this problem. Their application, however, can lead to a significant increase in system complexity and bit-rate. Alternatively, simple methods, e.g. fixed-length codes, can be used. Although their compression performance is lower, they are inherently synchronized; therefore, bit errors do not spread and corrupt entire image regions.
For browsing image databases, where images are primarily used for visualization purposes and fast reconstruction techniques are needed, depending upon the bandwidth, the combination of fixed-rate transmission methods and error concealment techniques might be preferable.
The initial work is based on a simple model which assumes that random errors are introduced in the bits that convey information about quantizer output levels. Assuming a binary-symmetric channel and random flippings to be statistically independent, an error concealment method has been proposed. This method simultaneously detects and corrects corrupted wavelet coefficients. It is observed that corrupted wavelet coefficients, i.e. coefficients whose codewords have been affected by random bit-flips, lead to characteristic error patterns which are, in fact, erroneous impulse responses associated with the filters of those channels where corruption occurred.