Matching patterns using Cross-Correlation | Essentials of ML

A visual explanation of how correlation is used to find objects in images or patterns in time series.
Also provides the explanation of different forms of correlation, connection to the convolution, and the convolutional neural networks.
This video is part of the Essentials of Machine Learning playlist:
• Let's make the Correla...
This tutorial is based on the notes by Professor David Jacobs.
Link:
www.cs.umd.edu...
#convolution
#deeplearning
#cnn

[@arpitmac]

I’m a little confused that when converting a 2d image to 1d how are the rows and column relations stay intact ? I am assuming it puts the each row into a vector but then for image kernels the values would be at a length of row difference from each other.

[@KapilSachdeva]

I am not converting 2d image to 1d image. 1d image or time series is just an example to illustrate the underlying operations. These operations work the same way on the 2d image as well, the difference is that for 2d image your kernel will be 2d also. If you have 2d kernel of size 3x3 then you will be doing cross-corrleation with 3x3 patch of your 2d image.

[@arpitmac]

@@KapilSachdeva that makes sense, I was just trying to confirm. Thank you!

[@KapilSachdeva]

🙏

[@younique9710]

Thank you for posting this great video! At 3:53, why did you use a "squared" Euclidean distance, instead of an Euclidean distance? I wonder if you use an Euclidean distance, the properties of the "squared" Euclidean distance are the same?