CAJ | 학술논문

传统的非正方形四叉树变换(NSQT)系数的扫描方式并不能很好地适应系数的能量分布，这将降低熵编码的编码效率。为了提高编码效率，该文提出非正方形四叉树变换系数能量分布模型，并改进了其系数扫描方式。首先推导了非正方形变换块的空间频率域位置分布，基于推导的分布，从理论上提出了其能量分布模型，并用实验进行了证实，发现能量从左上角到右下角逐渐递减，而左下角的能量与右上角的能量接近，根据该能量分布模型，重建了非正方形四叉树变换中量化系数的扫描模式，以适应熵编码，而熵编码中的系数扫描方式倾向于系数上的降序。实验结果显示，相比NSQT中已有的系数扫描方式，该文方法提出的扫描方式在编码计算复杂度和客观视频质量损失可忽略的前提下，平均下降了1.32%~3.74%的比特率，对于非正方形变换块，下降了9.05%~13.19%的比特率。
전통적비정방형사차수변환(NSQT)계수적소묘방식병불능흔호지괄응계수적능량분포，저장강저적편마적편마효솔。위료제고편마효솔，해문제출비정방형사차수변환계수능량분포모형，병개진료기계수소묘방식。수선추도료비정방형변환괴적공간빈솔역위치분포，기우추도적분포，종이론상제출료기능량분포모형，병용실험진행료증실，발현능량종좌상각도우하각축점체감，이좌하각적능량여우상각적능량접근，근거해능량분포모형，중건료비정방형사차수변환중양화계수적소묘모식，이괄응적편마，이적편마중적계수소묘방식경향우계수상적강서。실험결과현시，상비NSQT중이유적계수소묘방식，해문방법제출적소묘방식재편마계산복잡도화객관시빈질량손실가홀략적전제하，평균하강료1.32%~3.74%적비특솔，대우비정방형변환괴，하강료9.05%~13.19%적비특솔。
The conventional coefficients scanning pattern of the non-square quadtree transform can not fit the energy distribution better enough, which worsens the coding efficiency. In order to improve the coding efficiency, a non-square quadtree transform coefficients energy distribution model is proposed and the coefficients scanning pattern is improved in this paper. The position distribution in spatial frequency domain of the non-square quadtree transform block is deduced. Based on this distribution, the energy distribution model is proposed theoretically. The model is verified by experiment. It is found from the experiment result that the energy decrease gradually from the top-left corner to lower-right corner, and the energy in lower left corner is close to that in upper right corner. According to the energy distribution model, the quantized coefficients scanning pattern in non-square quadtree transform is rebuilt to fit the entropy coding, where the scanning pattern prefers to an ascending order in coefficients. Experiment results show that the proposed scanning pattern achieves average 1.32%~3.74% bitrate saving and 9.05%~13.19% bitrates saving for the non-square transformed blocks with negligible loss in coding computational complexity and objective video quality, compared with the scanning pattern in NSQT.