Teem: nrrd: Visible Human Female Color Data Processing

			Visible Human Female Color Data Processing
teem	/	nrrd		- Preliminaries - Slice Inspection - Cropping - Fiducial Inspection - Brightness Problem - Cube of Brain - Gray Scale Strips - Joint Histograms - Image Differences

The goal of these pages is three-fold:

Demonstrate how teem software can be used to inspect and analyze some of the RGB color data from the Visible Human Project. Specifically, demonstrate the utility of unu as a swiss-army-knife for processing raster data.
Describe the results of this analysis, especially concerning the problem of image brightness changing from slice to slice.
Document the steps taken in an effort to make a clean, "artifact"-free color volume dataset of the head of the Visible Female.

The quality of a visualization is related to the quality of the original data and the care with which it is processed. The goal here is to be as careful as possible when assembling a volume dataset from the RGB color images, and to learn as much as possible while doing so, so as to enable the best possible visualizations. This involves a few different tasks: inspecting the original RGB slices in an image viewer, cropping the images to include only the region of interest (the head), analyzing the changes in position of the fiducial markers to detect inter-slice translations, and finally analyzing (and trying to correct) the inter-slice variations in image brightness.

The software used for this is primarily a single command-line tool, unu, which, like cvs, is really multiple commands combined into one executable. Along the way, two other pieces of teem software are employed, an image manipulation program similar to mogrify called "ilk", and a tool for measuring values and derivatives in scalar and vector volumes, called "vprobe". Xv will be used to view PPM images, and in order to make PNG images for the web, I also use the convert program from ImageMagick. Using command-line tools may seem a little old-fashioned, but it means that if you have the Visible Human data, you can recreate 100% of what I did simply by copying and pasting from these pages into a shell. Every single image (except for one Matlab plot) appearing on the following pages can be regenerated exactly with the commands preceeding it.

		1) Preliminaries: Where to get the data, directory structure assumed by examples.
		2) Slice Inspection: How to look at one slice with `xv`, and/or downsample it. Demonstrates `unu make, permute, save, resample`.
		3) Cropping: Cropping the slices down the the interesting bits (the head, and two of the fiducial markers), and generating these for all head slices. Demonstrates `unu project, crop.`
		4) Fiducial Inspection: Looking at exactly how the fiducial markers moved from the top of the head to the bottom. Demonstrates `unu quantize, join, swap, 2op`.
		5) The inter-slice brightness problem: Using various kind of projections and histograms to visualize how the image brightness can change suddenly from slice to slice. Demonstrates `unu reshape, histax, heq, flip, rmap, slice`.
		6) Cube of Brain: A cubical region of the dataset is selected for closer scrutiny of the effects of the brightness problem. Demonstrates `vprobe.`
		7) Gray Scale Strips: Extracting the gray scale strips from the Kodak card in some of the images. Demonstrates `ilk.`
		8) Joint Histograms: Precisely visualize the brightness variations among small sets of images, and the unfortunate fact that the gray scale strips don't capture all the brightness variations inside the body. Demonstrates `unu jhisto.`
		9) Image differences: Per-pixel image differences also depict the brightness changes, as well as shedding a little light on why the gray scale strips failed.
???		10) Correcting the brightness problem. In progress.