• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.28 no.6C
• /
• pp.602-608
• /
• 2003

In this paper, we propose an efficient DCT calculation method for fast video encoding. We show that the SAD obtained in the motion estimation and compensation process is decomposed into the positive and negative terms. Based on a theoretical analysis, it is shown that the DCT calculation is classified into 4 cases - DCT Skip, Reduced_DCT1 , Reduced_DCT2, and original DCT- according to the positive and negative terms. In the proposed algorithm, one of 4 cases is used for DCT in order to reduce the computational complexity. The simulation results show that the proposed algorithm achieves computational saving approximately 25.2% without image degradation and computational overhead.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2019.11a
• /
• pp.249-251
• /
• 2019

In the ongoing standardization of Versatile Video Coding (VVC), DCT-2, DST-7 and DCT-8 are designated as the vital primary transform kernels. Due to the effectiveness of DST-4 and DCT-4 in smaller resolution sequences, DST-4 and DCT-4 transform kernel can also be used as the replacement of the DST-7 and DCT-8 transform kernel respectively. While storing all of those transform kernels, ROM memory storage is considered as the major issue. So, to deal with this scenario, a unified DST-3 based transform kernel derivation method is proposed in this paper. The transform kernels used in this paper is DCT-2, DST-4 and DCT-4 transform kernels. The proposed ROM memory required to store the matrix elements is 1368 bytes each of 8-bit precision.

• The KIPS Transactions:PartA
• /
• v.18A no.2
• /
• pp.39-44
• /
• 2011

This paper proposes the new high-performance circuit architecture of the transform and quantization for unified video CODEC. The proposed architecture can be applied to all kinds of transforms and quantizations for the video compression standards such as JPEG, MPEG-1/2/4, H.264 and VC-1. We defined the permutation matrices to reorder the transform matrix of the $8{\times}8$ DCT and partitioned the reordered $8{\times}8$ transform matrix into four $4{\times}4$ sub-matrices. The $8{\times}8$ DCT is performed by repeating the $4{\times}4$ DCT's based on the reordered and partitioned transform matrices. Since our circuit accepts the transform coefficients from the users, it can be extended very easily to cover any kind of DCT-based transforms for future standards. The multipliers in the DCT circuit are shared by the quantization circuit in order to minimize the circuit size. The quantization circuit is merged into the DCT circuit without any significant increase of circuit resources and processing time. We described the proposed DCT and quantization circuit at RTL, and verified its operation on FPGA board.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.16 no.12
• /
• pp.1236-1248
• /
• 1991

A new fast algorithm for computing the discrete cosine transform(DCT) Is developed decomposing N-point DCT into an N /2-point DCT and two N /4 point transforms(transpose of an N /4-point DCT. TN/t'and)It has an important characteristic that in this method, the roundoff noise power for a fixed point arithmetic can be reduced significantly with respect to the wellknown fast algorithms of Lee and Chen. since most coefficients for multiplication are distributed at the nodes close to the output and far from the input in the signal flow graph In addition, it also shows three other versions of factorization of DCT matrix with the same number of operations but with the different distributions of multiplication coefficients.

• Progress in Medical Physics
• /
• v.25 no.1
• /
• pp.53-63
• /
• 2014

The objective of this study is to investigate the difference of ITV lengths and ITVs between 4DCT and Slow-CT images according to respiratory patterns using a respiratory motion phantom. The respiratory periods 1~4 s and target motion 1~3 cm were applied on each respiratory pattern. 4DCT and Slow-CT images were acquired for 3 times. 4DCT and Slow-CT ITVs were measured with contouring the target in the Eclipse RTP system. The measured ITV lenghts and ITVs in 4DCT and Slow-CT images were compared to the known values. For the ITV lengths and ITVs in the 4DCT, the difference of them were reduced as the respiratory period is longer and target motion is shorter. For the Slow-CT, there was same tendency with change in 4DCT ITV lengths and ITVs about target motion. However, the difference of ITV lengths and ITVs for the respiratory periods were the lowest in respiratory period 1 second and different slightly within respiratory period 2-4 seconds. According to the respiratory patterns, pattern A had the highest reproducibility. Pattern B, C and D were showed the difference similar to each other. However, for pattern E, the reproducibility was the lowest compared with other four patterns. The difference of ITV lengths and ITVs between Slow-CT and 4DCT was increased by increasing the respiratory periods and target motion for all respiratory patterns. When the difference of Slow-CT ITV lengths and ITVs were compared with that of 4DCT ITV lengths and ITVs, Slow-CT ITV lengths and ITVs were approximately 22 % smaller than 4DCT, and the representations of target were different in each pattern. In case of pattern A, B and C, length difference was 3 mm at S (superior) and I (inferior) direction, and the length difference of pattern D was 1.45 cm at only "I" direction whereas the length difference of pattern E was 5 mm longer in "S" direction than "I" direction. Therefore, the margin in SI directions should be determined by considering the respiratory patterns when the margin of Slow-CT is compensated for 4DCT ITV lengths. Afterward, we think that the result of this study will be useful to analyze the ITV lengths and ITVs from the CT images on the basis of the patient respiratory signals.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.16 no.4
• /
• pp.765-770
• /
• 2012

In this paper, we proposed an effective hardware structure using DCT-based inra-prediction mode selection to reduce computational complexity caused by intra mode decision. In this hardware structure, the input block is transformed at first and then analyzed to determine its texture directional tendency. the complexity has solved by performing intra prediction in only predicted edge direction. $4{\times}4$ DCT is calculated in one cycle using Multitransform_PE and Inta_pred_PE calculates one prediction mode in two cycles. Experimental results show that the proposed Intra prediction encoding needs only 517 cycles for one macroblock encoding. This architecture improves the performance by about 17% than previous designs. For hardware implementation, the proposed intra prediction encoder is implemented using Verilog HDL and synthesized with Megnachip $0.18{\mu}m$ standard cell library. The synthesis results show that the proposed architecture can run at 125MHz.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 1998.06a
• /
• pp.121-126
• /
• 1998

DV(Digital Video) 영상 압축 방식에서 MPEG-2로 변환할 때 처리단계를 줄이기 위하여 DCT 영역에서 변환하였다. DV 방식의 색차신호 포맷인 4:1:1에서 4:2:2로 변환하고, 2-4-8 DCT 모드를 변환할 때 행렬을 이용하여 변환함으로써 중간과정을 줄였으며, DCT 영역에서 MPEG-2의 율 제어를 구현하였다. DV에서 만든 DCT 계수를 이용하여, 단계적으로 움직임 추정을 함으로써 전역탐색 블록 매칭 방식보다 처리 속도를 개선하였다.

• Journal of the Korean Institute of Telematics and Electronics C
• /
• v.35C no.5
• /
• pp.25-32
• /
• 1998

In this paper, SA (shape adaptive)-DCT is implemented using a datapath with 4 MACs (multiplication & accumulator). DCT is a well-known bottleneck of real-time video compression using MPEG-like schemes. High-speed pipelined MACs presented here implement real-time DCT. A datapath in this paper executes DCT/IDCT algorithms for QCIF 15fps(frame per second), maximum rate of VLBV(very low bitrte video) in MPEG-4. A 32bit accumulator in a MAC prevents distortion caused by fixed-point process. It can be applied to various operations such as ME (motion estimation) and MC(motion compensation) with a absolutor and a halfer.

• Proceedings of the Korean Information Science Society Conference
• /
• 2004.10c
• /
• pp.460-462
• /
• 2004

최신의 동영상 압축 표준인 H.264[1]는 기존의 동영상 압축 표준에 비해 압축 성능이 매우 높으며 4$\times$4 DCT(Discrete Cosine Transform)를 수행하는 특징이 있다. H.264 표준에서는 압축 효율을 높이기 위해 Intra 프레임 내의 이웃한 픽셀칸의 연관성을 이용한 프레임 내 창조(Intra Prediction)를 수행한다. 그러므로 기존의 동영상 압축 데이터를 H.264로 변환하기 위해서는 intra 프레임의 프레임 내 창조와 8$\times$8 DCT 블록의 4$\times$4 정수형 DCT 블록으로의 변환을 필수적으로 수행해야 한다. 또한, Intra 프레임은 GOP 내의 다른 프레임의 창조 대상이 되므로 변환 시 화질의 최적화가 필수적이다[2]. 본 논문에서는 Intra 프레임의 변환 시 화질의 최적화를 위해 DCT 도메인 상에서 프레임 내 창조를 수행하는 기법을 제안한다. 제안된 기법은 추가적인 계산없이 DCT 변환으로 인한 오류를 줄여 변환된 intra 프레임의 화질을 개선할 수 있다.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.25 no.3B
• /
• pp.541-548
• /
• 2000

In the MPEG-4 standardization phase, many methods for coding the irregular shaped VOP (video object Plane) have been researched. Texture coding is one of interesting research items in the MPEG-4. There are the Low pass extrapolation (LPE) padding, the shape adaptive DCT (SA-DCT), and the Extension-Interpolation (EI)/2D-DCT proposed in [1] as texture coding methods. the EI/2D-DCT is the method extending and interpolating luminance values from and Arbitrarily Shaped (AS) image segment into an 8 x 8 block and transforming the extended and interpolated luminance values by the 8x8 DCT. although the EI/2D-DCT and the SA-DCT work well in coding the As image segments. they are degraded since they use one-dimensional (1-D) methods such as the 1D-EI and the 1D-DCT in the two-dimensional (2-D) space. in this paper, we analyze the performance of the EI/2D-DCTand propose a new non-symmetric sig-sag scanning method, which non-symmetrically scans the quantized coefficients in the DCT domain to improve the EI/2D-DCT.