Old photos, old newspapers turn yellow with age. In the digital world, too, an image is ravaged by compression algorithms from the moment it is uploaded to the web, and ends up in front of us as a green image, which might look something like this:

Green accents, fuzzy details. Many people attribute these problems to the compression algorithm of Tieba. In fact, the problem is in the interface of compression image provided by Android before version 7.0. When RGB color space is converted to YUV, the image is repeatedly compressed with Skia’s defective color space transform algorithm, and becomes greener and greener.

The latter bug has been fixed, and in recent years, in addition to the old image, the green mesh image has also disappeared from our view.

Although the green picture is less, but in order to reduce the pressure on the server, the major app and website will also carry out image compression, image compression repeatedly, details will also be lost.

If you want to find original hd images on the web, you can use the image search engine to search for images. But in some cases, the original image is not clear, or has long been lost on the Internet, so superresolution technology is the only solution.

Artificial intelligence image enhancement kit

Waifu2x-extension-gui is one of the most powerful super frame completion software available today. It is open source on Github and currently has 3.4K stars.

Waifu2x himself is a Japanese technologist who came up with an image stretching algorithm after seeing a paper. Based on the operation of artificial intelligence convolution neural network, he reduced the pictures of a bunch of Gal and put their original pictures together, asking Waifu to learn how to enlarge and stretch.

Without further ado, let’s start with the effect. The original image is on the left, and it is recommended to zoom in to view it.

Before and after comparison of natural pictures:

The first time you use it, compatibility tests will be run automatically. There are multiple graphics engines built into the project, and there is always one that will work on your PC.

In addition, also support multithreading, multi – graphics card Settings. In one second, “fear of HD resets” was subdivided into “fear of ULTRA HD resets”.

I found a 156×162 low clarity bean, which is divided twice as much as the original one (left). You can see that much of the blur caused by the pressing image is gone, and the edges are smooth and clear:

Project address: github.com/AaronFeng75…