How do you group the components with same frequency into one channel?

[TongNing-Albert]

Exactly the detail process of DCT reshape process.

Thank you

[huangchaoxing]

Here is my interpretation with an example. If you get an image of 224224, and you would like to to divide them into 28 patches of a size of 88. For every patches, you apply the DCT and get an output of 8*8 (let's call it F matrix). Note that different element in the F matrix stands for different frequency. You may want to put the first element in the F1, F2... F28 into the first channel, and you may also want to put the Nth element in the F1, F2... F28 into the Nth channel. Thus you will get 28 by 28 by N by 3 tensor. The three here is for the three channels in the image domain.