A Subpixel Convolutional Image Super-Resolution Reconstruction Method Based on Multi-scale Labels

Abstract

The invention discloses a sub-pixel convolutional image super-resolution reconstruction method based on multi-scale labels, including: establishing and training a feature extraction structure, a residual module, an upsampling structure, a feature reconstruction structure and a cross-scale skip connection structure A sub-pixel convolutional network based on multi-scale labels is formed, and the network is used to complete the super-resolution reconstruction of the image. Convert the input image from RGB color space to YCbCr color space. Among them, the two channels of Cb and Cr use the bicubic interpolation and upsampling method to complete the super-resolution reconstruction work. The Y channel is sent to the trained network, and the super-resolution reconstructed image of the Y channel is output. The super-resolution reconstructed images of the Y, Cb, and Cr channels are fused to obtain the final high-resolution image. The present invention can quickly and accurately obtain super-resolution images, and the obtained super-resolution images can achieve good results in both subjective evaluation and objective image quality evaluation.

Description

technical field [0001] The invention belongs to the intersecting fields of digital image processing, deep learning and pattern recognition, and more specifically relates to a sub-pixel convolutional image super-resolution reconstruction method based on multi-scale labels. Background technique [0002] Improving the resolution of images is of great significance in the related fields of digital image processing. However, the image resolution is closely related to the image acquisition facilities, where the image sensor parameters and optical manufacturing technology determine the image resolution. Improving the hardware level of image equipment will bring huge economic costs. The image quality is also affected by uncontrollable factors such as the distance of the shooting environment, and the ability to solve the problem of image resolution by improving the shooting hardware level and the shooting environment is limited. [0003] In addition to improving the image resolution ...

AI Technical Summary

Problems solved by technology

[0004] In view of the above defects or improvement needs of the prior art, the present invention provides a sub-pixel convolutional image super-resolution reconstruction method based on multi-scale labels, thereby solving the problem of low accuracy of high-frequency image information prediction in the prior art, The problem of complex network structure and slow calculation speed

Images