摘要
改进单级字典学习的图像超分辨率算法,给出一种多级字典学习的图像超分辨率算法。通过多对字典的训练,记录不同层级退化图像和原始高分辨率图像之间的关系,由多对字典预测给定低分辨率图像不同层级丢失的高频信息,将预测出的高频信息与给定的低分辨率图像相加,得到逐级增强的高分辨率图像。在训练图像集相同的条件下,对于无噪声且没有压缩的低分辨率图像,改进算法相比单级字典学习的图像超分辨率算法,恢复出的高分辨率图像的峰值信噪比可平均提高约0.6dB。
The image super-resolution algorithm based on sigle-stage dictionaries learning is improved and an image super-resolution algorithm based on multi-level dictionaries is proposed.The relationship between different levels of the degraded image and the original high resolution images is recorded by training the multiple dictionaries.The multiple training dictionaries are used to predict the different levels high-frequency information lost in the given low resolution image.Add the predict high-frequency information to the given low resolution image and acquire the high resolution image with quality enhanced progressively.In conditions with the same training image set,for the noiseless and uncompressed low resolution images,the revised algorithm can recover high resolution images that the peak signal to noise ratio improves 0.6dB in average,compared to the image super-resolution algorithm based on single-stage dictionaries.
出处
《西安邮电大学学报》
2016年第3期32-37,共6页
Journal of Xi’an University of Posts and Telecommunications
基金
国家自然科学基金资助项目(61340040
61202183
61102095)
关键词
图像超分辨率
稀疏表示
字典训练
image super-resolution
sparse representation
dictionary training
