Efficient Rate Control and Fast Motion Estimation for Video Encoders

• Main Authors: Shih-Cheng Chang, 張世承
• Other Authors: Jef-Ferr Yang
• Format: Others
• Language: zh-TW
• Published: 2003
• Online Access: http://ndltd.ncl.edu.tw/handle/56668110564158306162

博士 === 國立成功大學 === 電機工程學系碩博士班 === 91 === In this thesis, we propose several techniques to improve video compression quality. First, human visual system (HVS) is included in rate control to improve the visual quality of reconstruct video sequences. A fast motion estimation is proposed to improve compression speed. At last, a predict rate control based on DCT indexes is applied to improve buffer fullness and frame skip situation. First, a new rate control scheme based on the human visual perception is proposed to enhance the visual quality for very low bit-rate video coding. The human visual perception developed in the DCT domain jointly considers both human visual systems and image activity measurements. Compared to the test model TMN8, which is developed under the optimal variance attribution to encoded bits, the proposed coding control scheme is designed for optimal visual power designation. Simulations demonstrate that the proposed coding control scheme achieves better visual quality in the same PSNR but encodes more frames than the TMN8 in the same bit-rate constraints. Second, another new rate control scheme based on human visual perception is proposed to enhance the visual image quality of very low bit-rate video transmission. With weighted DCT coefficients, the proposed rate control scheme can be easily designed with considerations of both the human visual system and skin enhancement in mind, to achieve a higher visual image quality for human perception. In comparison to the test model 8 (TMN8), which encodes the bits with optimal variance attribution, the proposed coding control scheme is designed to achieve the optimal visual power ascription. Simulation results demonstrate that the proposed coding control scheme encodes the same number of frames with nearly the same PSNR but achieves better visual quality than the TMN8 under the same bit rate constraints. Third, computation reduction algorithms of motion estimation are developed for low rate video coders. By jointly considering motion estimation, discrete cosine transform (DCT), and quantization, the all-zero and zero-motion detection algorithms are suggested to reduce the computation of motion estimation. Simulation results show that many unnecessary computations in motion estimation are greatly reduced. For Akiyo sequence in the QCIF format (176´144), the average number of the full search points is 787.88. The proposed algorithms, for examples, help three-step search and diamond search algorithms to reduce the average numbers of search points from 28.90 and 4.99 to 2.89 and 1.40 for each macroblock, respectively. Finally, we propose a model-based rate predictor, which is based on the indices of quantized DCT coefficients. The proposed rate predictor, which can characterize the detailed changes during encoding processing, provides more precise rate estimation than the traditional rate control methods, which are based on pixel variance. Based on the models of the proposed rate predictor, the optimal quantization step size can be easily obtained from a given bit rate constrain. Finally, the rate control mechanism based on the precise rate predictor is then developed. Built upon MPEG-4 video verification model (VM), our simulation results show that the proposed algorithm achieves superior rate control ability and improves PSNR of reconstructed video signal at the same time.