Computationally efficient wavelet decomposition/reconstruction for embedded systems

We propose an approach to the computationally efficient, wavelet based decomposition and reconstruction of digital images in embedded systems. This approach provides direct benefit in applications such as compression. Computational reduction is achieved by the parallel filter bank of the proposed outer product method, combined with the integerization of the coefficients, designed for embedded implementation. This paper discusses our architecture of the filter bank, how the integerization of the transform depends on this architecture, and how these integrized implementations improve performance in embedded systems. We also discuss our algorithm for creating integer transform approximations, defining what comprises optimal integer representations, giving convergence results.