We derive a numerical strategy for dimension reducing preprocessing to facilitate machine learning on high-dimensional imaging data. Certain orthogonal projectors are designed to reduce data dimension such that essential information content is preserved for subsequent learning. This strategy is applied to clinical spectral-domain optical coherence tomography data of the human retina for automated identification tasks.

This is poster number 20 in Poster Session

Authors:

Anna Breger (University of Vienna)

Keywords:

deep learning, dimension reduction, image deblurring, image enhancement, image representation, image segmentation, machine learning