Closed tanguyduval closed 8 years ago
Although its better now when using a discriminant analysis (we can keep more small axons), I think the best option is to do a different segmentation depending on the axon size. I can split my initial axon objects in 2 clusters (by optimizing the separation between the 2) & then perform all our next segmentation steps for both clusters (one at a time), including different discriminant analysis for each cluster. This is the optimal solution, but more complex (see screen captures below). So we end up having 2 segParameters mat files (one for each axon object domain) & at the end we combine these 2 segmentations to get our final segmentation. What do you think Tanguy?
I would recommend simplicity... Discriminant analysis should handle that. I close the issue
Large axons are too easily removed in circularity test. Test on: /Volumes/data_shared/microscopy/issues/i_20150623_rats_PANO_SEM/2%PFA2%Gluta_short_rince_CaCl2_PANO_SEM/20150823_PanoramaSEM/Section 4.png
Maybe create a new validation slider using MinorAxisLength and MajorAxisLength.