2D-mesh parallel structure of lifting wavelet transform
Yang Xiaoqiang · 2014
The hardware structures of lifting wavelet transform are mostly serial data processing, and processing time is very long. There are also some implementation of parallel processing for row transformation and column transformation, but row and column transform still is serial data conversion respectively, and not fully realize the parallel processing. In this paper, 2D-mesh parallel structure for lifting wavelet transform is proposed by taking full advantages of both regular data flow and local data properties of lifting wavelet transform. The structure realizes parallel wavelet transform, and the column does transform directly without data exchanging with memory after the row transform based the optimization of system structure and data flow. It increases the computational speed, and reduces the computational complexity and has good scalability. 2D-mesh structure proposed by the paper can reduce transform time to O(n) while transform time of the other proposed structures are normally O(n 2 ). The structure is adapt to VLSI implementation and applies to remote sensing image, video and the application with large amount of data and real-time processing of high performance requirements.