Abstract
This thesis is focused on low computational cost and low encoding delay algorithms for image/video coding and for the rate control problem. Ten algorithms and four estimates are proposed which can be classified along three main streams. In the first stream, three algorithms are designed for indirectly optimizing the visual quality for JPEG-LS for the same compression rates as the standard. They are based on the different tolerance of HVS across different textures and they exploit the prediction scheme of JPEG-LS for accurate texture identification. In the second stream, pixel trend variation is used in the design of two algorithms in order to improve the compression rates by up to 9% in the padding problem of boundary macroblocks in MPEG-4. Pixel trend variation is also used through two novel algorithms in order to reduce the MSE error by up to 3% for the same compression rates in the case of diagonal edge detection inJPEG-LS. In the third stream, four novel estimates are proposed to optimize "on-the-fly" the RD performance of the MPEG-2 rate control scheme. This is achievable through exponential modulation, non-variance based activity estimates and local activity estimation. Improvements by up to 3.5db for the same bit rates as the standard were observed. Finally, two algorithms that
improve the "rate only" MPEG-2 rate control scheme in R-D terms, by explicitly considering distortion in the quantization step size assignment through a Lagrangian based local optimization are proposed. The proposed schemes achieve improvements of 0.5-ldb as compared to the Idb improvements of traditional R-D optimization techniques for the same compression rates as the standard but with just a small fraction of the operations needed. All the algorithms and the estimates proposed are ideal for online compression applications where low computational complexity and encoding delay are of prime importance.
| Date of Award | Mar 2001 |
|---|---|
| Original language | English |
| Supervisor | Mike Reddy (Supervisor) |