HEVC와 JPEG 하드웨어 부호화기를 위한 DCT의 Approximate Calculation

학위논문 (석사)-- 서울대학교 대학원 : 전기·컴퓨터공학부, 2015. 8. 이혁재.Discrete Cosine Transform (DCT) is widely used for various image and video compression applications because of its excellent energy compaction property. DCT is computationally intensive and the calculations are parallelizable. Therefore it is often implemented in hardware for speeding up the calculation. However due to large size of DCT or multiple modules of DCT required for some applications, the hardware area taken up by DCT in image or video encoders become significant. The DCT required in most applications doesnt need to be exact. Taking advantage of this fact, here a novel approach is provided to reduce the hardware area cost of the DCT module. The DCT hardware module consists of combinational logic and memory. Both the components are reduced and the complete implementation is described. The application being aimed at is for HEVC and JPEG, however the idea is applicable to any DCT hardware implementation. Finally the degradation caused to encoded image and video in terms of BDBR is discussed and the gate count results from the synthesis is provided.Chapter 1 Introduction 1 1.1 2D DCT Hardware Module . . . . . . . . . . . . . . . . . . . . . 2 1.1.1 Pipelining the process . . . . . . . . . . . . . . . . . . . . 5 1.2 Approximate DCT . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Chapter 2 Related Works 9 Chapter 3 The Moving Window Idea for Bit-Width Reduction 12 3.1 ML Recovery for Moving Window . . . . . . . . . . . . . . . . . 16 Chapter 4 Approximate DCT for HEVC 19 4.1 HEVC Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 4.2 HEVC Encoder . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 4.3 DCT in HEVC Encoder . . . . . . . . . . . . . . . . . . . . . . . 21 4.4 Approximate DCT in HEVC . . . . . . . . . . . . . . . . . . . . 23 4.4.1 The three components of the DCT module . . . . . . . . 27 4.4.2 Optimizing Partial Butterfly Adder/Subtractors . . . . . 29 4.4.3 Optimizing the multiplication module . . . . . . . . . . . 30 4.4.3.1 Multiple Constant Multiplication (MCM) . . . . 32 4.4.3.2 Approximate MCM . . . . . . . . . . . . . . . . 32 4.4.4 Optimizing the transpose memory . . . . . . . . . . . . . 36 Chapter 5 Approximate DCT for JPEG 39 5.1 JPEG Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 5.2 Approximate DCT . . . . . . . . . . . . . . . . . . . . . . . . . . 41 5.3 Application of Moving Window to DCT transpose memory . . . 42 5.3.1 Ideal implementation . . . . . . . . . . . . . . . . . . . . . 43 5.3.2 Window position based on first row . . . . . . . . . . . . . 43 5.3.2.1 Cases of failure . . . . . . . . . . . . . . . . . . . 46 5.3.3 Position based on first column . . . . . . . . . . . . . . . 48 5.3.3.1 Cases of failure . . . . . . . . . . . . . . . . . . . 49 5.4 Hybrid implementation . . . . . . . . . . . . . . . . . . . . . . . . 50 Chapter 6 Experimental Results 54 6.1 HEVC Experiments and Results . . . . . . . . . . . . . . . . . . 55 6.2 JPEG Experiments and Results . . . . . . . . . . . . . . . . . . . 55 Chapter 7 Conclusion 64Maste