You are viewing the course site for a past offering of this course. The current offering may be found here.
Lecture 22: Image Processing (13)
tanmayghai18

The discrete cosine transform used here for JPEG's is actually quite similar to the DFT many of us have learned before, just using only real numbers. In fact, they are equivalent to DFTs of roughly twice the length, on data twice the length of those used for DFT.

youtuyy

In regard to DFT and DCT: basically, DCT is used for those processes where low-frequency content will be emphasized. Such as in speech or image coding. For spectral analysis purposes, DFT would yield a better tool.

There is also FFT, which is a collection of algorithms for fast computation of the DFT. Typically the number of operations required by the FFT is on the order of N*logN.

frankieeder

I don't believe it was mentioned in lecture but the coefficients in the upper left corner are the most important for human visual perception. As such, quantization matrices often try to preserve information in the upper left coefficients and zero out the rest.

tyleryath

Check out this video by one of my favorite youtube channels to get a better understanding of DCT: https://www.youtube.com/watch?v=Q2aEzeMDHMA

You must be enrolled in the course to comment