1 / 15

A Zero-Value Prediction Technique for Fast DCT Computation

A Zero-Value Prediction Technique for Fast DCT Computation. Y. Nishida, K. Inoue, and V. Moshnyaga. Introduction. Discrete Cosine Transform (DCT) is the most commonly used transform for video coding (JPEG, MPEG H26x etc). Problem. DCT is a very computationally expensive task.

lave
Download Presentation

A Zero-Value Prediction Technique for Fast DCT Computation

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. A Zero-Value Prediction Technique for Fast DCT Computation Y. Nishida, K. Inoue, and V. Moshnyaga

  2. Introduction • Discrete Cosine Transform (DCT) is the most commonly used transform for video coding (JPEG, MPEG H26x etc). Problem • DCT is a very computationally expensive task. • DCT requires a measurable amount of energy and many operations. Reducing the number of operations and energy consumption of DCT becomes important. 2

  3. Our Proposal A Zero-Value Prediction Technique for Fast DCT Computation Advantage • Reduces the number of DCT and quantization operations. • Reduces energy consumption. • Increases DCT operation speed. Disadvantage • The picture quality deteriorates. 3

  4. The feature of DCT The high frequency components at the DCT output are concentrated nearby zero value. DCT Input data (Pixel Level) Output data (DCT coefficient) 4

  5. The feature of Quantization The probability of elements in the encoding block to become zero after quantization is very high. Quantization Input data (DCT coefficient) Quantization output data In the case Quantization step = 16 5

  6. Zero-Value Prediction Technique When the DCT output N zeros, we predict the remaining result is also zero without actual DCT computing. In the case with N=2 The detected string of zeros The predicted value Starting point of zero-value prediction Reducing operations for 35 DCT coefficients is possible. 6

  7. A Zero-Value Prediction Technique Example (1/2) DCT coefficient Conventional Technique A Zero-Value Prediction technique The prediction errorincreases but most of the DCT coefficients will be zeros after quantization. 7

  8. A Zero-Value Prediction Technique Example (2/2) Quantization output data Conventional Technique A Zero-Value Prediction technique Prediction mistake Prediction mistake When a picture is decoded, the prediction error degrades image quality 8

  9. Evaluation Using the MPEG software from MPEG Software Simulation Group • 2D-DCT algorithm use addition 1024 and multiplication 1024 times per 1 block. • 2D-DCT use partitionable method (1D-DCT×1D-DCT). • Raster scan order of DCT computation. Video benchmarks missa carpohne foreman salesman QCIF CIF QCIF QCIF 150frame 382frame 298rame 300rame 9

  10. PSNR as a function of N   carphone foreman missa salesman × ▲ 70 60 50 PSNR (dB) 40 30 20 10 0 3 1 2 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 N • The PSNR is low when N is small. • The PSNR does not change a lot after N=9. 10

  11. Operation reduction rate as a function of N(DCT)   carphone foreman missa salesman × ▲ Operation reduction rate 0.5 0.4 0.3 0.2 0.1 0 3 1 2 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 N • The reduction rate is high when N is small. 11

  12. Operation reduction rate as a function of N(quantization)   carphone foreman missa salesman × ▲ Operation reduction rate 1.0 0.8 0.6 0.4 0.2 0 3 1 2 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 N • The reduction rate is high when N is small (similarly to the DCT operation). 12

  13. Example of reconstructed picture for N=9 Assumption toaccept N=9 as a tradeoff for both picture quality and the reduction rate. The example of a picture Original method Zero-value prediction(N=9) The PNSR decreases by 1.07dB. 48.75dB 47.67dB 13

  14. Conclusion A Zero-Value Prediction Technique for Fast DCT Computation • When N=9, the number of computations is reduced by 29% for DCT and by 59% for quantization. • The average PSNR value degreases by -1.6dB. Future work • Estimating the effect of the proposed technique on energy consumption. 14

  15. Comparison of the algorithm The number of operations per 1block. Normal : addition 1024 and multiplication 1024 times. Chen : addition 416and multiplication 224 times. FFT : addition 464and multiplication 176 times. 15

More Related