Computational complexity reduction methods for multiscale recurrent pattern algorithms

Abstract

The Multidimensional Multiscale Parser algorithm was originally proposed as a generic lossy data compression algorithm. An high degree of adaptivity and versatility allowed it to outperform state-of-the-art transform-based compression methods for a wide range of applications, from still images, compound documents, or even ECG's, just to name a few. However, as other pattern matching algorithms, it presents a high computational complexity. In this paper, we investigated several techniques that allowed to considerably reduce both the encoder's and the decoder's computational complexity, with marginal R-D performance losses. The most important reduction was achieved on the decoder, that reduced up to 95% the time required by the previous method. These improvements contribute to affirm MMP as an alternative to traditional transform-based encoders, approaching its computational complexity with that of transform-based algorithms.