基于SVR和PCA的超分辨率图像恢复算法应用研究

利用单幅低分辨率图像重构超分辨率图像的算法中,通常基于样本库进行图像重构,而这类算法效率较低。提出了一种利用SVR和PCA进行特征压缩的图像重构算法,其基本思路是将训练图像分解成若干个基本小块作为样本库;然后利用PCA对低分辨率图像基本小块进行降维处理,并将得到的主成分系数作为特征加以训练,在识别和重构过程中,将待恢复图像进行回归分析,找到相应的超分辨率图像块,然后进行重构。实验结果表明,本文方法较其他算法有更优的恢复结果,并能同时保证较好的实时处理特性,很好地逼近了原始的真实图像。

Super-resolution image restoration algorithm based on SVR and PCA

Using a single low resolution image to reconstruct a super-resolution image, usually based on the sample image reconstruction, but this kind of algorithm efficiency is low. This paper presented a SVR and PCA based image restoration method. Firstly, it decomposed the low resolution images into several small pieces and projected these samples onto a smaller space. Then trained the SVR using these samples and their corresponding high resolution patches. During the restoration procedure, it decomposed the test image in the same way and project using trained PCA model. After SVR, mapped each low resolution patch to a high resolution patch which was used to restore the final image. The experiments show that the method can achieve better performance than cubic interpolation method and also has very high computational efficiency.