摘要
针对视频序列中逐帧图像的超分辨率复原问题,提出一种基于参考高分辨率图像的视频序列超分辨率复原算法.该算法利用了大多视频设备能同时提供静止图像和动态视频拍摄的特点,以同一场景获取的高分辨率静止图像为参考图像,提取高频细节经运动估计补偿和可信度加权后用于低分辨率序列各帧的超分辨率复原,并采用最大后验概率约束优化进一步提高复原图像的保真度.实验中采用多个序列对算法性能进行了测试,结果表明,该方法对各序列中连续多帧的复原效果均明显优于传统的双线性插值方法和基于最大后验概率(MAP)的静态批处理方法,其平均PSNR值与MAP静态批处理方法相比提高了2.4 dB以上.
To solve the problem of sequence to sequence super resolution, a video sequence super resolution algorithm based on a frame of reference high resolution image is proposed in this paper. The algorithm is based on the fact that in most cases, the vidicon device has dual mode： HR（High Resolution） still and LR（Low Resolution） video. In the algorithm, a HR still image acquired from the same scene is used as an example, the high frequency details of it are extracted and then motion compensated and reliability weighted for super resolution of the corresponding low resolution frames. The estimated HR frames are then constraint optimized under the Maximum A Posteriori（MAP） framework to improve its fidelity. Evaluated by several sequences, the proposed algorithm shows obvious performance improvements compared with the traditional bilinear interpolation and the MAP based static batch method. The average PSNR（Peak Signal-to-Noise Ratio） of the reconstructed frames are improved by more than 2.4 dB.
出处
《北京工业大学学报》
CAS
CSCD
北大核心
2009年第6期742-747,共6页
Journal of Beijing University of Technology
基金
国家自然科学基金资助项目(60472036
60772069)
北京市自然科学基金资助项目(4052007)
北京市科技新星计划基金资助项目(2005B08)
关键词
图像序列
图像分辨率
运动补偿
image sequence
image resolution
motion compensation
