Error resilient lossless compression based on binary RLE

Abstract

The billion pictures shot on consumer cameras or computer generated are mostly stored in compressed data format, often vulnerable even to a single bit change. Obviously, the file format used for storage is an issue, especially at the end-of-life of the storage medium, when hardware become fragile and migration is necessary, increasing the probability of being exposed to bit errors while reading, transmitting and re-writing data.Error resiliency was a guideline for the JPEG2000 standard [9] and [5], besides this, some formats have already been designed for lossless compression (PNG, JPEG-LS, JPEG2000, BMF, TMW, LOCO-I, CALIC, …). However only few academic works target efficient lossless compression combined with error resiliency. This early approach points only luminance image, and aim to reduce local redundancies to approaching Shannon's entropy. First, an adaptative predictor based on a causal neighborhood is used to estimate the value of the current pixel, and the error signal is subsequently processed, up to run-length encoding of bitplanes using bit-patterns to store the length of the runs. At many stages of the processing, the impact of a possible bit error is minimized and after run-length encoding, the data is protected with an EDC/ECC (Error Detection Code / Error Correction Code) using log2(N) metadata for that usage. Last, a bit shuffle (exactly a Bit-Reversal Permutation[6]) is applied for protection against burst errors. Results will be shown and compared to JPEG2000 images.