Exploring approximations in 4- and 8- point DTT hardware architectures for low-power image compression

Abstract

Due to the intensive use of discrete transforms in picture coding, the search for fast and power-efficient approaches for their hardware implementation gains importance. The Discrete Tchebichef Transform (DTT) represents a discrete class of the Chebyshev orthogonal polynomials, and it is an alternative for the Discrete Cosine Transform, commonly used in picture coding. High energy compaction and decorrelation are the main properties of the DTT. The state-of-the-art approximate DTT matrix is composed of 0, 1, − 1, 2, and − 2 coefficient values. In this work, we propose a new approximation for both the 4-point and 8-point integer DTT with better quality and power-efficiency. We explore the effects of coefficient truncation, whose values are 1/16, − 1/16, 1/8, − 1/8, 1/4, and − 1/4. Considering operations with integers, the smaller values of coefficients causes truncation in the internal transform calculations and leads to smaller values for the non-diagonal residues, which reduces the non-orthogonality. We have also selectively pruned the rows of the state-of-the-art approximate DTT matrix. The results show that the proposed pruned approximate DTT hardwired solutions increases the maximum frequency up to 5%, minimizes circuit area by over 30%, with savings of up to 32.4% in power dissipation with a higher compression ratio and fewer quality losses in the compressed image, when compared with state-of-the-art approximate DTT hardware designs.