Large Image Weka

Hi everyone! I am trying to analyze images taken of FISH samples of rat brain tissue. All of my images are between 150 and 200 MB. Even with my computer’s max available 25 GB RAM allocated to ImageJ, I get a Memory error when I use the same Weka Segmentation Feature Stack as I did on smaller images. However, when I run with no image features selected, I am able to get a complete segmentation, albeit a rather poor one. My question is this: Which image features would be best for me to use in the Trainable Weka Segmentation tool given my memory constraint and the image sample I provide below?

Here is a segment of one of the much larger images:

Hello @lbmcp,

One option is use less features indeed. Given the complexity of your image, I suggest you try selecting only Gaussian and optionally Difference of Gaussian.

Another option would be to split your large images into smaller images, process them and merge them back together.

I hope this helps!



Apart from selecting less training features, you can as well reduce the number of trees.

1 Like