Issue No. 02 - Feb. (2018 vol. 67)
Haider A.F. Almurib , Faculty of Engineering, The University of Nottingham, Semenyih, Malaysia
Thulasiraman Nandha Kumar , Faculty of Engineering, The University of Nottingham, Semenyih, Malaysia
Fabrizio Lombardi , Department of ECE, Northeastern University, Boston, MA
This paper proposes a new framework for digital image processing; it relies on inexact computing to address some of the challenges associated with the discrete cosine transform (DCT) compression. The proposed framework has three levels of processing; the first level uses approximate DCT for image compressing to eliminate all computational intensive floating-point multiplications and executing the DCT processing by integer additions and in some cases logical right/left shifts. The second level further reduces the amount of data (from the first level) that need to be processed by filtering those frequencies that cannot be detected by human senses. Finally, to reduce power consumption and delay, the third level introduces circuit level inexact adders to compute the DCT. For assessment, a set of standardized images are compressed using the proposed three-level framework. Different figures of merits (such as energy consumption, delay, power-signal-to-noise-ratio, average-difference, and absolute-maximum-difference) are compared to existing compression methods; an error analysis is also pursued confirming the simulation results. Results show very good improvements in reduction for energy and delay, while maintaining acceptable accuracy levels for image processing applications.
Discrete cosine transforms, Image coding, Adders, Approximation algorithms, Algorithm design and analysis, Delays, Signal processing algorithms
H. A. Almurib, T. N. Kumar and F. Lombardi, "Approximate DCT Image Compression Using Inexact Computing," in IEEE Transactions on Computers, vol. 67, no. 2, pp. 149-159, 2018.