Create Super-Resolution Images

Open in Cloud
Download Notebook

Apply sharp upsampling to images using neural nets

Get the pre-trained network from the Wolfram Neural Net Repository:

In[1]:=

net=NetModel["Very Deep Net for Super-Resolution"]

Out[1]=

NetChain



Input port:	image
Output port:	array (size: 255×255×1)



This network performs the refinement on the Y channel in the YCbCr color space. Write an evaluation function to handle net resizing and color conversion:

In[2]:=

netevaluate[img_,imgScale_]:=Block{interpolated,ycbcr,channels,resizedNet,diff},interpolated=ImageResize[img,Scaled[imgScale],Resampling"Cubic"];ycbcr=

RGBtoYCbCr

@interpolated;resizedNet=NetReplacePartnet,"Input"

NetEncoder[

]

;channels=ColorSeparate[ycbcr];diff=Image[resizedNet[First[channels]]];ycbcr=ColorCombine[channels+{diff,0,0}];

YCbCrtoRGB

@ycbcr

Start with a small image:

In[3]:=

img=

;

Evaluate the network on the image to double its size:

In[4]:=

refined=netevaluate[img,2]

Out[4]=

Compare with the unrefined resampling using the same cubic kernel used to produce the net input and a high-quality OMOMS kernel:

In[5]:=

upsampledCubic=ImageResize[img,Scaled[2],Resampling"Cubic"];upsampledOMOMS=ImageResize[img,Scaled[2],Resampling{"OMOMS",7}];TableFormImageImageTrim[#,{{{200,20},{380,200}},{{200,200},{300,300}}}],

...