I am about to develop a multi-purposed API for image and video processing with my team. I am searching the internet to find some resources (youtube, blog, medium articles, papers, github, etc) to answer these questions below:
1- why do we convert images to frequencies? (what are the advantages of this action?)
2- In convolutional neural network, in what circumstances we transform images to frequencies? are there any specific advantages involved there also?
3- Does CNN understand the features of image better than the normal image?
It is highly appreciated if anyone could help me with resources (specially github codes for projects which use CNN trained model on image converted frequencies) .
Thanks
When you convert image to frequency domain, you generally apply some type of transform such as Fourier or discrete cosine transform. The objective some time is to perform image compression or some types of operations are more convenient in frequency domain. Remember that convolution in spatial domain is multiplication in frequency domain.
The following research paper might be of help for you.
Noman Islam
Thanks for tips
I'm sorry but I don't see any referenced paper in your answer. Have you forgot to add the link Noman?
Masoud Masoumi Moghadam This is a free resource from cognitive-ai - https://courses.cognitiveclass.ai/courses/course-v1:CognitiveClass+ML0120ENv2+2018/courseware/76d637cbe8024e509dc445df847e6c3a/d2529e01786a412fb009daef4b002a48/.
It includes codes which can be useful to develop your model. I have used it and hope it helps.
- Badges
- Science topic
- Similar topics
- Computer Science
- Images