Performance analysis of a multiple-image super-resolution implemented on a mobile device

College

College of Computer Studies

Document Type

Conference Proceeding

Source Title

DLSU Research Congress 2017

Publication Date

2017

Abstract

Multiple·image super-resolution (SR) attempts to recover a high·resolution image (HR) from a set of low·resolution (LR) images. Multiple·image super·resolution restores high·frequency details and produces an HR image by utilizing pixel differences obtained from multiple LR images. Processing of multiple images and attempting to produce an HR image can be computationally expensive. Thus, SR systems are normally implemented on desktop PCs. However, mobile devices already have capable hardware and mobile image processing libraries, such as OpenCV, are available. A multiple-image SR system .:an therefore be implemented on a mobile device if available resources are carefully managed. To prove this claim, a prototype application was implemented on an Android device. This paper discusses the system architecture of the mobile multiple·image SR system which consists of the following modules: Input Module, Edge Detection Module, Denoising Module, Image Alignment Module, Alignment Selection Module, and Image Fusion Module. In terms of performance time, the Denoising Module has the longest average processing time of 3 minutes and 28 seconds. The Image Alignment Module, Alignment Selection Module and Image Fusion Module has an average processing time of 1 minute each. In terms of memory consumption, the Image Fusion Module consumes the largest amount of memory of 3S(R) where S is the scaling factor and R is the resolution size of the input LR images. A total of 3 matrices are required to perform the image fusion operation. Optimization techniques sucli as matrix pooling and thread barriers are also discussed.

html

Disciplines

Computer Sciences

Keywords

High resolution imaging; Imaging systems—Image quality

Upload File

wf_no

This document is currently not available here.

Share

COinS