Image Processing?

[sethjuarez]

Image Processing?

[bdschrisk]

Hey, at the moment numl isn't really suited for image processing. You would ideally use a convolutional neural network for recognising specific parts in patient label images. However, if you were to convert the image to text first using OCR (check out the Microsoft portable OCR library) and then apply a Naïve Bayes or Logistic Regression for recognising race, etc., then numl would work.

[sethjuarez]

I would love to add an Image feature (even if just to pass off the numbers to a better algorithm). My guess is we would resize all the images on the first pass then use rgb number as the actual value (then normalize). It would be fun to implement... right?? 😄

[bdschrisk]

Seth, I think we should look into this soon.

We need to revisit the Matrix library to include depth for each of the colour channels. We then need to create sampling algorithms i.e. Convolutional and Pooling nodes / layers in the Neural Network objects.