A Look-Up-Based Universal Real-Time Transformer for Image Coding

A universal transformer for two-dimensional separable linear transforms in image coding is proposed. It is not restricted to fast transforms, and switching between transforms only requires writing the appropriate constants into an internal RAM. Look-up tables are used to reduce computational complexity. The linear transform properties are exploited to reduce the memory requirements by decomposing the input vector into smaller arrays and into bit planes. The input is pipelined into parallel data paths. An implementation example on a single gate array with existing technology achieves real-time transformation of 8 × 8 blocks, with area requirements of around 38K gates.