Manifold Projection Image Segmentation (MPIS)

I created an image segmentation pipeline that handled 10% noise and reduced analysis time by 85% for a 25k-pixel 75D spectral image. This improvement leveraged encoding spectra and then segmenting the image using these encodings (rather than using the spectra themselves). The reordering reduced sensitivity to noise and allowed incorporation of additional information, such as positional encoding and other spectroscopic measurements, all of which increased accuracy.