Scalable low-complexity B-spline discrete wavelet transform architecture

A scalable discrete wavelet transform architecture based on the B-spline factorisation is presented. In particular, it is shown that several wavelet filters of practical interest have a common structure in the distributed part of their B-spline factorisation. This common structure is effectively exploited to achieve scalability and to save multipliers compared with a direct polyphase B-spline implementation. Since the proposed solution is more robust to coefficient quantisation than direct polyphase B-spline, it features further complexity reduction. Synthesis results are reported for a 130-nm CMOS technology to enable accurate comparison with other implementations. Moreover, the performance of the new wavelet transform architecture, integrated in a complete JPEG2000 model, has been collected for several images.