Abstract

Main idea: We present a novel software framework for image processing tasks that combines packing of input bitplanes in order to process multiple increments of the input together and increment previously-computed results. The resulting framework for incremental computation is validated using transform decompositions, 2D convolution and cross-correlation, and frame-by-frame block matching.

Key results: The execution time results with bitplane-based computation indicate that the proposed approach can be comparable or superior to conventional (non incremental) computation for several cases. We have also investigated results under real-time scheduling and were able to demonstrate that, by exploiting the incremental nature of the proposed computation, the worst-case (“digital world”) approach of: ‘Can this image processing task be performed in X frames-per-second?’ changes to the best-effort (“analog world”) approach of: ‘What is the achieved quality when this task is performed in X frames-per-second?’. Similarly, we present power-distortion results demonstrating that the proposed software-based incremental computation allows for seamless prolongation of the battery life of a low-power device with a simple change of output quality level. Finally an interesting application for region-of-interest computation is presented.

Further research: Even though the proposed framework deals with linear or quadratic operations, we discuss the possibility of application to other (non-linear) types operations, which is non-trivial and requires further research.Papers and software: http://www.ee.ucl.ac.uk/~iandreop/ORIP.html

Bio: Yiannis Andreopoulos received the Ph.D. degree in applied sciences from the Vrije Universiteit Brussel, Brussels, Belgium, in 2005, as well as the M.Sc. degree in signal and image processing systems and the Electrical Engineering Diploma, both from the University of Patras, Patras, Greece.

He is presently Lecturer at the Department of Electronic and Electrical Engineering, University College London, London, U.K. His research interests are in the domain of multimedia signal processing.