University of Utah              The JOHN A. MORAN EYE CENTER

HOME

TOOLS

SCIENCE

PUBLICATIONS

CYTOSOMICS & METABOLOMICS

IMAGE CLASSIFICATION TEST SETS

Registered 8-bit tiff series

2546_small_cropped_inv_work_masked_d.TIF 2546_small_cropped_inv_work_masked_e.TIF 2546_small_cropped_inv_work_masked_g.TIF 2546_small_cropped_inv_work_masked_q.TIF 2546_small_cropped_inv_work_masked_r.TIF 2546_small_cropped_inv_work_masked_y.TIF

Registered 8-bit tiff binary (0-255) cell object mask

2546_small_masked_lo7_idl_thememask.tif

Registered 8-bit indexed Isodata classification result

2546_small_masked_lo7_idl_themeindexed.tif

Registered 8-bit Jewels Test Set

Marc_2000_jewels_512_R.tif Marc_2000_jewels_512_G.tif Marc_2000_jewels_512_B.tif Marc_2000_jewels_512_RGB.tif

Registered 8-bit grey 2366 GC Test Set: All images 1350 x 964

Marc_2002_chan_b.tif Marc_2002_chan_d.tif Marc_2002_chan_e.tif Marc_2002_chan_q.tif Marc_2002_chan_t.tif Marc_2002_chan_y.tif Theme maps Marc_2002_2366map.tif: Refined theme map with kerf removed and classes 5 & 8 separated by deconvolution Marc_2002_gc_indexed_map.tif: Original raw theme map produced by PCI Geomatica using Isoclus with max classes=20, target classes=16, min classes=12, sample=20%, threshold=1%

Classification "Noise" viewed in parallel plots with CellKit A parallel-plot algorithm reveals two "variance" sources our "refined" map for cell classes: intrinsic biological variance and outliers likely due to map misalignment. The plot is a polyline for each 7D point, connecting its partners on 7 parallel axes, for each class. Class 01 - noise = a sparse band of horizontal lines outside the main "M" shaped plot. SD=15 for class 01, channel 0 (chan_b). Class 2 - noise =a heavy band of horizontal lines ( a lot of bad pixels included under the mask) outside the main tilted "M" shaped plot. But SD is still only 8 for channel 0 (chan_b) because the density of "good" pixels is really high. Class 3 - very few bad pixels, but the intrinsic dispersion is a little higher. (channel 0 SD =13) Obviously, the next thing to do is write an explore tool to highlight coordinates of the "bad" pixels in the polyline on the original image.

HOME

TOOLS

SCIENCE

PUBLICATIONS

CYTOSOMICS & METABOLOMICS