基于CNN的图像超分辨率重建方法

Image super-resolution reconstruction based on CNN

为同时满足较好的超分辨率重建效果和实时处理要求,提出一种基于CNN的图像超分辨率重建方法。以低分辨率图像作为输入,采用1×1的小卷积核进行降维和扩维处理,减少网络的参数,利用反卷积与池化的组合提取出对结果更为敏锐的特征,通过反卷积进行上采样重建,易于实现图像不同比例放大。实验结果表明,相比FSRCNN-s、ESPCN等方法,该方法实现了更好的重建效果,平均每秒能处理24张以上尺寸为320×240的图像,满足对视频超分辨率重建的实时性要求。

To satisfy better super-resolution reconstruction effects and real-time processing requirements at the same time, an image super-resolution reconstruction method based on convolutional neural network was proposed. The low-resolution image was used as input, and the 1×1 small convolution kernel was used for dimensionality reduction and expansion to reduce the network parameters, the combination of deconvolution and pooling was used to extract features that were more sensitive to the results, deconvolution was used for upsampling and reconstruction, and it was easy to achieve different scales of image magnification. Experimental results show that compared with FSRCNN-s, ESPCN and other methods, the proposed method not only achieves better reconstruction effects, but processes more than 24 images with a size of 320×240 per second, which meet the real -time requirements for video super-resolution reconstruction.