CORDIC based novel energy-efficient approximate DCT architecture for error-resilient applications

Abstract

There is energy conflict in battery operated multimedia devices that leads to poor and degraded image/ video quality. In such type of devices the most crucial, vital and computationally exhaustive compression standard element used is the Discrete Cosine Transform (DCT). This paper instigates one such novel DCT architecture that employs very well-known COordinate Rotation DIgital Computer (CORDIC) unit and some approximation techniques which provides with effective computation based on certain assumptions where the correlation between the nearby pixels is exploited and dependency between the coefficients is used. The proposed architecture transpires out to have just 16 coefficients as the end result and not all of the 64 coefficients. This amalgamates the interdependence amid the DCT coefficients (a, b to g) that in turn squeezes them into three (seven to three), thereby allowing the computation of the other four from these three using adders and shifters. The effectiveness of the proposed architecture is evaluated by using various parameters like Peak Signal to Noise Ratio (PSNR), Structural Similarity Index (SSIM), Gradient Mean Square Deviation (GMSD) that are available for evaluation of image quality via modeling them in MATLAB. The architecture turns out to be nearly 54%, 30%, 86% and 55% more efficient in terms of energy over the radix-2 CORDIC, CORDIC based Loeffler DCT, data priority DCT and the RICO based DCT architectures.