In real-time video streaming applications based on battery-powered camera systems, it is crucial to capture and transmit critical information satisfying the memory capacity and channel bandwidth requirement. However, the increase of data rate in ultra high denition video contents aggravates the memory capacity and the channel bandwidth problem. Therefore, one of the crucial issues is to minimize the bit-rate of the video encoding so as to satisfy the system constraints. Most of the rate control schemes has the large frame-to-frame distortion variation at the cost of regulating the bit-rate, thereby it mostly leads to poor perceptual video quality according to the human visual system. In this dissertation, I present two single-pass bit allocation schemes: i) frame-level bit allocation to minimize the frame-to-frame distortion variation satisfying the bit-rate constraint, and ii) coding tree unit (CTU)-level bit allocation scheme to minimize the encoding bit-rate while satisfying keeping the perceptual video quality. In Chapter 1, I propose a single-pass content-aware video encoding scheme to achieve a consistent video quality while satisfying the bit-rate constraints. To achieve the objective I developed i) scene descriptor based on the discrete cosine transformation (DCT) coefficients obtained from the video encoder, ii) a parameterized analytic distortion model for video encoding using the scene descriptor, iii) scene change detection based on the encoded macroblock (MB) mode ratio, and iv) a single-pass consistent video quality control scheme which exploits i), ii), and iii). In chapter 2, I propose a CTU-level bit allocation scheme to achieve the minimization of encoding bit-rate considering the perceptual video quality of the reconstructed frames after the video encoding process. To achieve the objective, I developed: i) an analytical model for the structure similarity (SSIM) index estimation based on the statistics of the input frame only; a zero quantization parameter as an indicator for the decision of the zero block encoding; and iii) a CTU-level bit allocation scheme that exploits i) and ii). Experimental results are very promising compared to the previous methods.
이 논문에서는 실시간 영상 스트리밍에서 비디오 인코딩의 비트 율-왜곡 제어를 위한 최적화된 비트 할당
방법을 다루었다. 좀 더 자세히 말하면, 비디오 인코딩 시 영상 내의 프레임간 왜곡의 변화를 최소화하여
일정한 영상 화질을 제공하기 위한 프레임 레벨 비트 할당 방안과 비트 율 최소화를 위한 코딩 트리 유닛
레벨 비트율 할방 방안을 제시하고 그 성능을 살펴보았다. 제안한 비트 할당 방법은 이제까지의 다른 방법들과 견주어 더 나은 성능을 갖으며, 실시간 영상 스트리밍을 위한 유망한 방법이 될 수 있다.