Basic image feature extraction

[agitter]

As an extension to our neural network feature extraction (#5), we could do a high-throughput HTCondor run to extract basic image properties like:

• entropy per channel
• total intensity per channel

Then once we have the clustering in #5, we could see whether any of those are strongly associated with cluster membership. The overall goal is still to identify any obvious image attributes (e.g. empty images) that split the images into the two major clusters.

[agitter]

Recall from https://github.com/xiaohk/pharmaco-image/blob/master/meta_data.ipynb that many of these summary statistics, like intensity, have already been computed.