| 摘要: | 隨著多媒體影像解析度需求提高，高畫質多媒體介面HDMI (High
Definition Multimedia Interface)解析度規格也往上提升，傳輸頻寬需求也隨
之提高，儼然已觸及物理傳輸上限，為解此問題，VESA (Video Electronics
Standards Association)協會推出DSC (Display Stream Compression)影像串流
壓縮標準，適用於數位影音介面傳輸標準的影像壓縮。
VESA DSC 具有視覺無損(Virtually Lossless)與低延遲(Low Latency)的
特性，為滿足視覺無損，VESA DSC 壓縮率範圍在2 至5 倍之間，低壓縮
率既可維持畫質的同時，也可符合物理傳輸的上限。另外為滿足低延遲，
VESA DSC 的運算不能過於複雜，故採用空間壓縮(Spatial Compression)，
可有效減少運算與儲存空間需求。
在VESA DSC 架構中，本研究提出一種線性預測(Linear Prediction)方
法，在不改變其預測演算法則公式的結構下，額外新增預測時的參考點，
平均可降低0.49%的BDBR。而在碼率控制(Rate Control)上，由於每個slice
第1 條線預測效果較差，本研究提出一種在每個Slice 第1 條線做額外位
元分配(Bit allocation)，找出最佳分配位元數，平均可降低0.3%的BDBR。
最後同時整合線性預測與碼率控制改進，平均BDBR 降幅進一步提升至
0.78%。;As the resolution requirements of multimedia images increase, the
resolution specifications of High Definition Multimedia Interface (HDMI) are
also rising, leading to a corresponding increase in transmission bandwidth
demands, seemingly reaching the physical transmission limit. To address this
issue, the VESA (Video Electronics Standards Association) introduced the
Display Stream Compression (DSC) standard, applicable to image compression
in digital audio-visual interface transmission standards.
VESA DSC features visually lossless and low latency. To achieve visually
lossless, the VESA DSC compression ratio ranges from 2 to 5 times. The low
compression ratio maintains image quality while meeting the physical
transmission limit. Furthermore, to achieve low latency, VESA DSC′s
computation cannot be overly complex; therefore, it employs spatial
compression, effectively reducing computational and storage space requirements.
In the VESA DSC architecture, this study proposes a linear prediction
method. Without changing the structure of the prediction algorithm, it adds additional reference points during prediction, resulting in an average reduction of 0.49% in BDBR. Regarding rate control, since the prediction performance of
the first line of each slice is poor, this study proposes an additional bit allocation
for the first line of each slice to find the optimal number of allocated bits,
resulting in an average reduction of 0.3% in BDBR. Finally, by integrating
improvements in linear prediction and rate control, the average BDBR reduction
was further increased to 0.78%. |
