Fast DWT-based intermediate video codec optimized for massively parallel architecture

摘要

A method and apparatus for scalable symmetric data compression and decompression (codec), optimized for real-time processing of high-resolution continuous-tone 2D input, such as video and images. Following a dyadic n-level DWT decomposition of input, the coding of subbands is spatially partitioned to enable strong parallelism that maps into concurrent encoding threads, based on typical artifacts of H-bands on initial levels—large areas of contiguous zeros interspersed with sparse and highly redundant non-zero values. Variable length codewords of both non-zero values and their positions are mapped into a linear bitstream by a combined limited sequential and parallel procedure of at most log₂ n iterations, with both lossless and lossy variants enabled. The compressed bitstream of self-delimiting entries is decoded in a symmetric iterative procedure that delivers to decoding threads the non-zero values and position codewords for final image restoration.