ORIP - Operational Refinement of Image Processing - Sponsor: EPSRC

This is the repository for papers and software related to progressive (or incremental) computation of image processing tasks. Our efforts are separated into hardware-oriented ORIP, software-oriented ORIP and operational tight packing (OpTiP). OpTiP is a tool used for concurrent arithmetic operations in software, which is essential for software-oriented ORIP.

• Hardware-oriented ORIP

• Software-oriented ORIP

• OpTiP

• Downloads

• Demonstrators

Hardware-oriented ORIP

The following paper is an approach to compute the Harris salient point detector progressively. This detector is the cornerstone of virtually all gradient-based detection algorithms today. The computational effort vs detection accuracy under the assumption of sequential adder and multiplier hardware is examined in detail in the paper:

• Y. Andreopoulos and I. Patras, "Incremental refinement of image salient-point detection," IEEE Trans. on Image Processing, vol. 17, no. 9, pp. 1685-1699, Sept. 2008.

The following two papers describe the incremental computation of the discrete wavelet transform under the direct 2D lifting scheme algorithm. The second paper presents results for the energy consumption of the arithmetic operations on a Xilinx FPGA implementation:

• Y. Andreopoulos and M. van der Schaar, "Incremental refinement of computation for the discrete wavelet transform," IEEE Trans. on Signal Processing, vol. 56, no. 1, pp. 140-157, Jan. 2008.
• Y. Andreopoulos, D. Jiang and A. Demosthenous, "Prediction-based incremental refinement for binomially-factorized discrete wavelet transforms," IEEE Trans. on Signal Processing, vol. 58, no. 8, pp. 4441-4447, Aug. 2010.

Software-oriented ORIP

For a general (less technical) overview of software-oriented ORIP see:

• D. Anastasia and Y. Andreopoulos, "Operational Refinement of Image Processing: ORIPv1.0" London Communications Symposium, paper 48, Sept. 2009.

For the short (conference) paper with the core technical concepts see:

• D. Anastasia and Y. Andreopoulos, “Software designs of image processing tasks with incremental refinement of computation,” Proc. IEEE Workshop on Signal Process. Systems (SIPS), Tampere, Finland, Oct. 2009.

For a longer conference paper with emphasis on application results see:

• D. Anastasia and Y. Andreopoulos, “Scheduling and energy-distortion tradeoffs with operational refinement of image processing ,” Proc. Design Automation and Test in Europe (DATE), Dresden, Germany, Mar. 2010.

For the full journal paper with description, analysis and applications, see:

• D. Anastasia and Y. Andreopoulos, “Software designs of image processing tasks with incremental refinement of computation,” IEEE Trans. on Image Processing, vol. 19, no. 8, pp. 1057-7149, Aug. 2010.

NEW!!! Throughput-distortion computation of generic matrix multiplication in software:

• D. Anastasia and Y. Andreopoulos, "Throughput-distortion computation of generic matrix multiplication: toward a computation channel for digital signal procesing systems," IEEE Trans. on Signal Processing, to appear.

NEW!!! Throughput scaling of convolution for error-tolerant multimedia applications:

• M. A. Anam and Y. Andreopoulos, "Throughput scaling of convolution for error-tolerant multimedia applications," IEEE Trans. on Multimedia, to appear.

OpTiP - Operational Tight Packing

OpTiP is a framework for the establishment and usage of maximum packing in the calculation of a linear processing task by a computer. For the full paper with the analysis and experiments, see:

• D. Anastasia and Y. Andreopoulos, "Linear image processing operations with operational tight packing," IEEE Signal Processing Letters, vol. 17, no. 4, pp. 375-378, Apr. 2010.

Download all projects now from HERE!

These software packages are licensed under a Creative Commons Attribution-Non-Commercial-Share Alike 2.0 UK: England & Wales License. Licenses for commercial usage can be obtained from UCL, please contact Y. Andreopoulos.

Demonstrators  

• At SIPS'09, Davide Anastasia demonstrates incremental refinement of a filtering application versus conventional computation, both under scheduling jitter. The conventional approach leaves significant pars of the calculation incomplete, while the incremental approach provides a "best effort" result. These are indicated as "incomplete (or uncovered) frames" and "striped (or not fully-completed) frames"

• The two demos at the University Booth of DATE 2010: (i) A conventional laptop and webcam running frame capturing and incremental processing under scheduling deadlines; (ii) The OLPC ("100$ laptop") running frame capturing and incremental edge detection with quality/framerate tradeoffs

This page last modified 17 January, 2012 by [Yiannis Andreopoulos]