基于L0范数稀疏表达的图像盲超分辨率重建

发布时间：2018-11-19 17:56

【摘要】：由于大部分超分辨率图像重建方法都是建立在图像的点扩散函数为已知或假设点扩散函数为高斯模糊核的条件下,但真实的低分辨率图像中的点扩散函数并不是高斯函数,而是由随机的相机抖动造成的。为了提高重建的超分辨率图像质量并使其更接近真实场景,提出了一种基于L0范数稀疏表达的图像盲超分辨率重建方法。首先利用了基于L0范数的梯度最小化方法估计出超分辨率图像中的点扩散函数,再通过点扩散函数的估计在超分辨率重建的过程中有效地去除图片的模糊效应,最后利用反向传播算法,使重建的超分辨率图像更接近真实。通过实验结果表明,提出的方法相对于双三次插值法和基于多字典学习的图像超分辨率重建算法可以得到更清晰的重建效果,峰值信噪比和平均结构相似度均有提高,最后在真实图片重建测试效果中也得到了更好的验证。
[Abstract]:Because most of the super-resolution image reconstruction methods are based on the condition that the point diffusion function of the image is known or assumed to be Gao Si fuzzy kernel, but the point diffusion function in the real low-resolution image is not Gao Si function. It is caused by random camera jitter. In order to improve the quality of the reconstructed super-resolution image and make it closer to the real scene, an image blind super-resolution reconstruction method based on L0 norm sparse representation is proposed. Firstly, the gradient minimization method based on L0 norm is used to estimate the point diffusion function in the super-resolution image, and then the fuzzy effect of the image is effectively removed during the super-resolution reconstruction by the estimation of the point diffusion function. Finally, using the back propagation algorithm, the reconstructed super-resolution image is closer to reality. The experimental results show that the proposed method can obtain a clearer reconstruction effect than the bicubic interpolation method and the image super-resolution reconstruction algorithm based on multi-dictionary learning, and the peak signal-to-noise ratio (PSNR) and the average structural similarity are improved. Finally, in the real image reconstruction test results are also better verified.
【作者单位】：河南工程学院计算机学院;郑州大学信息工程学院;
【基金】：国家自然科学基金(61379079)
【分类号】：TP391.41

【相似文献】