Fast VLSI overlapped transform kernel

Transforms of image sequences are commonly compositions of discrete cosine/sine transforms. The advantage is that the N-point DCT/DST requires only Nlog(N) multiplications when properly decomposed in the well known butterfly structure. The butterfly decomposition removes redundancy in the transform. This provides speed-up but not so much cost reduction because numerical sensitivity sets a price on the implementation. An alternative way is to guarantee robustness, by relying on orthogonal arithmetic, and exploiting this robustness to make computations inexpensive and, therefore, transformations fast. This concept and its merits are the subjects of this paper. An example from image coding is given.