基于图像分辨率增强算法的场景生成技术  

Virtual Scene Genaration Technology Based on Algorithm of Image Resolution Enhancement

作　　者：赵艳伟[1] 

机构地区：[1]吉林工商学院信息工程分院,吉林长春130062

出　　处：《液晶与显示》2011年第6期823-830,共8页Chinese Journal of Liquid Crystals and Displays

基　　金：吉林省教育厅"十一五"科研规划项目(No.2011222;2011224);吉林省科技厅自然科学基金项目(No.201105064)

摘　　要：针对高分辨率真实感的虚拟环境及场景浏览时变焦观察的需要,研究了基于图像的高分辨率场景生成技术。图像分辨率增强技术是指利用已采样的信息来重新构建分辨率更高的场景图像,包含单帧图像的分辨率增强和多帧图像序列的分辨率增强两种技术。对于单帧图像的分辨率增强技术,提出了一种基于熵变分的图像分辨率增强算法。该算法在贝叶斯估计和最大熵原理的基础上,将图像像素点梯度信息应用到图像分辨率增强中,从而建立起一种基于图像梯度信息的各向异性自适应分辨率增强算法。对于多帧图像序列的超分辨率复原技术,在单帧熵变分模型的基础上,将双边滤波技术引入到图像超分辨率复原中,建立了一种基于广义熵变分的图像超分辨率复原模型,提出了一种基于几何距离和梯度信息的双重加权各向异性分辨率增强算法。实验结果表明:使用本文算法得到的高分辨率复原图像具有较高的峰值信噪比和视觉质量,与传统图像分辨率增强算法相比具有一定的优势。The scene generation technology based on algorithm of image resolution enhancement,aiming at the needs of high image resolution and zooming browse in virtual environment was studied.The algorithm of image resolution enhancement contains two situations.First,for one frame,a novel image zooming algorithm based on maximum a posteriori probability and maximum entropy principle was presented,which anisotropy interpolate pixels using gradient information self-adaptively.Second,in the case of multi-frames,super-resolution reconstruction algorithm based on generalized MAP theory and maximum entropy principle was presented.By introducing bifilter into traditional MAP algorithm,we anisotropy interpolate pixels using gradient information and geometry distance self-adaptively.Experimental results show that our algorithm holds higher signal/noise ratio and better vision quality.

关 键 词：高分辨率 单帧图像 多帧图像序列 双边滤波 信噪比 

分 类 号：TP911[自动化与计算机技术]