SegmentImage

Segments a directory of images with various methods. The method choices are:

Constant method thresholds the image intensities with the given value. Image intensities go from 0 to 1.
Otsu's method thresholds the image intensities so that the intra-class variance is maximized. The calculated threshold value can be multiplied with a correction constant, to improve the result.
Shape method thresholds the image at several intensity levels and searches for objects fitting the size and roundness criteria given in the parameters.
ConstantArea method thresholds the image at several intensity levels and returns a mask where a certain ratio of image area is included in the mask. Filling/size restrictions are applied after masking. Use constant to control the area ratio. A minimization function is used, so the area fraction may not be accurate.
KMeans method thresholds the image intensities by performing k-means algorithm on the intensities, and by sorting the clusters in cluster center order.
Snake method, or Active Contour method, uses an iterative method to find edges from an initial guess. Uses ellipses as seeds for finding the actual objects. Algorithm implementation copyright (c) 2010, Ritwik Kumar, All rights reserved. BSD License.
Correlation method thresholds based on two grayscale images, where threshold level lies on the linear regression curve, and the correlation coefficient for pixels below the treshold is less than 0.
Wavelet method uses the ATrousWaveletBlobDetection function developed by Dirk Padfield, GE Global Research. Use opt input for optional background mask (white for bg, black for fg). Reference: Padfield, D., Rittscher, J., and Roysam, B., "Coupled minimum-cost flow cell tracking for high-throughput quantitative analysis", Medical Image Analysis 15(4), 650-668 (2011)
Grid method creates a grid of constantxconstant pixels, with a spacing of 1 pixel around each square.

Example:

Source	Outputs

Chart for the minsize/maxsize area restrictions:

Chart for the minround/maxround roundness restrictions:

Version	2.3
Bundle	anima
Categories	Image Processing
Authors	Ville Rantanen (ville.rantanen@helsinki.fi)
Issue tracker	View/Report issues
Requires	Matlab
Source files	component.xml shapesegment.m imagesegment.m kmeanssegment.m snakesegment.m snakeiterate.m correlationsegment.m execute.m
Usage	Example with default values SegmentImage( in = ImageList, //opt = ImageList, //seed = MaskList, clearBorders = true, clusters = 2, constant = 0.5, corr = 1, fill = true, maskList = false, maxArea = 0, maxEcc = 1, maxRound = 1, method = "Otsu", minArea = 0, minEcc = 0, minInt = 0, minRound = 0, numbers = false, snakeParam = "0.7,0.65,1,0.7,0.65,0.65,0.65,100", waveletParam = "4,5" )

Inputs

Name	Type	Mandatory	Description
in	ImageList	Mandatory	A directory with grayscale image files.
opt	ImageList	Optional	A directory with grayscale image files used in Correlation method.
seed	MaskList	Optional	Masklist of objects as initial guess for the 'Snake' method. e.g. output from Ellipse2Masklist.

Outputs

Name	Type	Description
mask	ImageList	BW PNG images.
perimeter	ImageList	BW PNG images of the perimeter of the mask.
numbered	ImageList	BW PNG images of the numbers of the objects in the mask. Created only if parameter numbers is set true.
masklist	MaskList	Image folder structure describing each object. One folder of images for each source image.

Parameters

Name	Type	Default	Description
clearBorders	boolean	true	Clear objects touching the image edges. Applicable to all methods.
clusters	float	2	Clusters for KMeans method. The result mask will have as many gray scales as cluster number. Results with number higher than 2 will not be compatible with the feature measurements components, and object numbering doesn't work.
constant	float	0.5	Constant value for threshdolding the images, for 'Constant' method. Range 0 to 1. Also, the area ratio in ConstantArea method.
corr	float	1	Correction multiplier for threshold value in 'Otsu' and 'Correlation' method.
fill	boolean	true	Fill holes in segmented objects. Applicable to all methods.
maskList	boolean	false	Produce masklist entry for each object. Forced true with 'Snake' method.
maxArea	float	0	Maximum accepted pixel area in all methods. 0-1 is considered as ratio of image size. Use 0 for no limit.
maxEcc	float	1	Maximum accepted eccentricity value. Default=line
maxRound	float	1	Maximum accepted roundness value.
method	string	"Otsu"	Method for image segmentation. Choose from 'Otsu', 'Constant', 'Shape', 'ConstantArea', 'KMeans', 'Snake', 'Wavelet', 'Correlation', 'Grid'.
minArea	float	0	Minimum accepted pixel area in all methods.
minEcc	float	0	Minimum accepted eccentricity value. Default=circle
minInt	float	0	Minimum accepted intensity value in 'Shape' and 'Otsu' method. Ranges 0-1.
minRound	float	0	Minimum accepted roundness value.
numbers	boolean	false	Draw images of the numbers of the objects in the mask. This can be a lengthy process if there are huge amounts of objects.
snakeParam	string	"0.7,0.65,1,0.7,0.65,0.65,0.65,100"	Parameters for active contour method. Must be a comma separated list of eight numbers. Explanation in the following order: alpha: Controls tension beta: Controls rigidity gamma: Step size (note, smaller is bigger step) kappa: Controls enegry term wl: line component weight we: edge component weight wt: terminal energy component weight iterations: No. of iteration for which snake is to be moved.
waveletParam	string	"4,5"	Parameters for Wavelet method. Must be a comma separated list of two numbers. Explanation in the following order: minscale: Smalles scale level to use maxscale: Largest scale level to use

Test cases

Test case	Parameters▼	IN in	IN opt	IN seed	OUT mask	OUT perimeter	OUT numbered	OUT masklist
case1	properties	in	(missing)	(missing)	mask	perimeter	(missing)	(missing)
	method=Shape, minRound=0.6, maxRound=1, minArea=200, maxArea=2000
case2	properties	in	(missing)	(missing)	mask	perimeter	(missing)	(missing)
	method=Otsu, minArea=200, maxArea=2000, corr=0.7, numbers=true, maskList=true
case3_kmeans	properties	in	(missing)	(missing)	mask	perimeter	(missing)	(missing)
	method=KMeans, clusters=3, numbers=false, minArea=30
case4_snake	properties	in	(missing)	seed	mask	perimeter	numbered	(missing)
	method=snake, numbers=true, minArea=10, snakeParam=0.1, 0.35,0.007,0.4,0.25,0.65,0.25,40

Generated 2019-02-08 07:42:07 by Anduril 2.0.0