A real time, low latency, FPGA implementation of the 2-D discrete wavelet transformation for streaming image applications

In this paper, we present an architecture and a hardware implementation of the 2-D Discrete Wavelet Transformation (DWT) for applications where row-based raw image data is streamed in at high bandwidths and local buffering of the entire image is not feasible. The architecture is especially suited for multi-spectral imager systems, such as on board an imaging satellite, however can be used in any application where time to next image constraints require real-time processing of multiple images. The latency that is introduced as the images stream through the DWT filter and the amount of locally stored image data is a function of the image and tile size. For an n(1) x n(2) size image processed using (n(1)/k(1)) x (n(2)/k(2)) sized tiles the latency is equal to the time elapsed to accumulate a (1/k(1)) portion of one image. In addition, a (2/k(1)) portion of each image is buffered locally. The proposed hardware has been implemented on an FPGA and is part of a JPEG2000 compression system designed as a payload for a Low Earth Orbit (LEO) micro-satellite, which will be launched in August 2003.