基于兴趣区的编码方法探讨

Discussison on ROI-based Coding Method

目的探讨JPEG2000标准中基于兴趣区(region of interest,ROI)的编码方法,重点讨论了标准中的最大上移(maximum shift,maxshift)方法。方法Maxshift方法缩放ROI的系数,使得属于ROI的所有信息在码流中放在背景区(back-ground,BG)的信息之前。改变图像的量化方式和码流截断点的位置,就能获得不同质量的ROI和BG。结果在低比特率(大约小于0.05bpp)时,ROI和BG具有近似的图像质量。随着比特率的增加,ROI的质量迅速提高。结论Maxshift方法处理任意形状的ROI时,不需要编码形状,解码器既不用解码形状,也不需要生成ROI掩膜,因此更加简单。

Objective To discuss the encoding methods based on ROI in JPEG2000 still image coding standard and pose emphasis on Maxshift method. Methods Maxshift method is based on scaling of ROI coefficients so that all the information pertaining to ROI is placed before the information pertaining to BG. By varying the quantization of the image and by truncation of the code stream, different quality for ROI and for the background can be obtained. Results ROI and BG have very similar image quality at low bitrates （approximately lower than 0.05bpp）,but with the increase of bitrates, the quality of ROI improves much faster. The quality of ROI is significantly better than that of the image encoded without ROI, particularly at low bitrates. Conclusion the maxshift method is simpler without the need of shape coding at encoder and without decoding shape and ROI mask generation at the decoder when dealing with arbitrarily shaped regions of interest.