ApproxVision: Approximating the Image By Exploiting the Limitations of Human Visual System

Buswala, Dinesh and Mittal, Sparsh (2018) ApproxVision: Approximating the Image By Exploiting the Limitations of Human Visual System. Masters thesis, Indian Institute of Technology Hyderabad.

Preview

Text
Thesis_Mtech_CS_4168.pdf - Published Version
Download (3MB) | Preview

Abstract

Approximate computing has recently emerged as a promising approach to the energy-efficient design of digital systems. Approximate computing relies on the ability of many systems and applications to tolerate some loss of quality or optimality in the computed result for saving energy and performance enhancement. In image processing, applications impose high energy consumption in loading and accessing the image data in the memory. Fortunately, most image processing applications can tolerate approximation in processing. The quality of service (QoS) of image processing applications depends upon the human visual system. The Human Visual system has some limitations like weak peripheral vision and not able to distinguish the difference between the quality of the original image and processed image when PSNR value is greater than 30 dB. These limitations give us a hint that instead of approximating the entire image we should take the peripheral part of the image because the human eye has the lower peripheral vision and high center vision and at the same time processed image has PSNR value greater 30 dB to gain the excellent quality of an image. Leveraging these facts we proposed one approximate computing technique that will save energy without sacrificing the QoS. We will approximate only the peripheral part of the image, and in the peripheral region, we change lower bits in each pixel because the contribution of lower bits in a pixel is less compare to higher order bits in a pixel. The proposed technique will take care of the limitations of the human visual system to approximate the images.

[error in script]

IITH Creators: