dreamplayerzhang/superstitchous2.0

Image stitching and core segmentation software for large scale digital holography.

#Superstitchous 2.1

Purpose: Image stitching and core segmentation software for large scale digital holography (Quantitative Phase Imaging). You must cite this work by providing a reference to this page.

Authors: Kevin Han, Zelun Luo

Mentor: Mikhail Kandel

Requirement:

• Visual Studio 2013 (Visual Studio 2012 will not work)

• Python 3

  • Extension package: NumPy, SciPy, PIL

• OpenCV 3.0 alpha (OpenCV 2.x will not work)

• Boost 1.56.0

• fftw (provided)

• libtiff (provided)

• Eigen (provided)

• tminres (provided)

• CMake for GCC (YMMV)

Usage: For stitching and core segmentation, run

python wrapper.py [-s <config file for stitch>] [-c <config file for seg>]

Output of stitching is in a format suitable for uploading to CATMAID or [Zoomify] (http://www.zoomify.com).

Additionally, TIFF files are provided for inspection in ImageJ.

The config file is JSON, consisting of a single object with these keys:

###Config file parameters for wrapper.py: ####Stitching:

Parameter	Description
inDir	directory with poslist file; file name must end in "poslist"
outDir	output directory
projName	project name for
zoomLvls	number of zoom levels

Optional parameter	Description	Default
pixelRatio	Conversion from stage coordinates to pixels	1 pixel/micron (off)
maxPeakX	Max pixel offset of phase correlation peak from stage value for horizontally offset images.	dx / w * 0.2 sqrt(w*h), where dx is x overlap.
maxPeakY	Max pixel offset of phase correlation peak from stage value for vertically offset images.	dy / h * 0.2 sqrt(w*h), where dy is y overlap.
maxPeakXY	Max pixel offset of phase correlation peak from stage value for diagonally offset images.	(dx*dy) / (w*h) * 0.2 sqrt(w*h)
cacheSz	# images to cache when reading.	50
poslist	Path of poslist relative to inDir.	Any file ending in "poslist" (can only be one)
skipAlign	true to skip aligment, and raster the "poslist" as-is	false
weightPwr	During least squares optimization, the error for each offset is weighted by peak sum ^ weightPwr. So, 0 = no weighting.	1
peakRadius	Amount on each side of the peak to sum up. 0 = 1 pixel total (just the peak); 1 = 9 pixels total, etc.	1
fixNaN	true to replace NaN values in image with zeros.	true
bgSub	percentage of input images used to generate the background. (0 if not doing background subtraction)	0
outWidth	Rasterized tile width	1024
outHeight	Rasterized tile height	1024
writeJPG	Write JPGs?	true
writeTIF	Write floating point TIF files?	false
colormapMin	Color for pure black (0,0,0)	-0.3
colormapMax	Color for pure white (1,1,1)	0.8
rasterFormat	Naming convention for files, choose from CATMAID or ZOOMIFY	CATMAID

The defaults for peak offset linearly extrapolate between 0% image overlap (0 peak offset; trust stage completely), and 100% overlap (peak offset is 0.2 * "image dimension"). (Maybe this will work?)

####Segmentation:

Parameter	Description
inDir	directory with project.yaml; output directory of stitching.
outDir	output directory of segmented cores.
coreW	number of cores in each row.
coreH	number of cores in each column.

Optional parameter	Description	Default
winW	width of segmenting window in cropped.jpg.	55
winH	height of segmenting window in cropped.jpg.	55
crop	1 = generate the output; 0 = only view the segmentation result in downsampled image.	0