Does CNN work on figures without patterns

AI Discussions
1

Hello Everyone,
Just a quick question and would like to hear your opinions.

For some applications, mostly biomedical applications, we may want to build a classifier based on the color information, not the patterns.

For example, suppose we are monitoring some biosamples for a long period and would like to develop a classify based on the long-term monitored images. The patterns in the images are kept the same and are not useful. The only difference is from the color change of some pixels (randomly distributed) caused by the bioprocesses.

Does CNN work on these applications? I got a CNN working but I am curious why? Since there are no useful patterns in the images.

Any ideas would be helpful.

Regards,

Yangyang Zhao

2

What is a pattern? How do measure differences in them? What is a color? How do you measure differences in them? Seems like if you can define those, you can put them into an objective/cost/loss calculation and use it to drive optimization/learning.