What are the features extracted from the images when using linear regression?

Feature requests
#1

Hi, I am doing a thesis on object detection and I used the linear regression as learning block. One of my panelists required me to provide the full feature extraction step-by-step process. Can you guys help me out what features are extracted? Is it color, edge, etc… Thank you!

Convert image into raw features
#2

Hello @niwlyer,

Here you can have a look at how our pre-processing blocks work: https://github.com/edgeimpulse/processing-blocks

For the image pre-processing block, we are only using the number of channels and the size of the image as variables.
If you want to go further in the image pre-processing, I worte an article some time ago on how to add your own custom DSP block that works with images: https://www.edgeimpulse.com/blog/utilize-custom-processing-blocks-in-your-image-ml-pipelines

Note that it won’t work with our Object Detection pre-trained models but you can use this with a custom DL model.

Regards,

Louis

1 Like
#3

Hi Louis,

I am new to this topic. When linear regression is applied, what feature does it take from the image? Color, shape, etc, or the whole image?

Thank you for the response! :smile:

#4

Hi @niwlyer,

Ii will take all pixels of the image, either as RGB or Grayscale depending on your Image block configuration.

Aurelien

#5

Hi @aurel ,

Appreciate the response. Do you have a post/article about the pre-processing steps for the Linear Regression (Keras) on the learning blocks. Learning the specific steps will help alot.

Thank you!

#6

Hi @niwlyer,

You can use any processing block with the linear regression. You can read more about our official blocks here, you can also create your own custom blocks following the documentation.

Aurelien

#7

Thank you for the help @aurel