regiongrowing (Operator)

Name

regiongrowing — Segment an image using regiongrowing.

Synopsis

regiongrowing(Image : Regions : Row, Column, Tolerance, MinSize : )

Description

regiongrowing segments images into regions of the same intensity --- rastered into rectangles of size Row * Column. In order to decide whether two adjacent rectangles belong to the same region only the gray value of their center points is used. If the gray value difference is less then or equal to Tolerance the rectangles are merged into one region.

If g_{1} and g_{2} are two gray values to be examined, they are merged into the same region if:

    |g_1 - g_2| < Tolerance

For images of type 'cyclic', the following formulas are used:

          |g1 - g2| < Tolerance and |g1 - g2| <= 127
    256 - |g1 - g2| < Tolerance and |g1 - g2| >  127

For rectangles larger than one pixel, ususally the images should be smoothed with a lowpass filter with a size of at least Row * Column before calling regiongrowing (so that the gray values at the centers of the regtangles are “representative” for the whole rectangle). If the image contains little noise and the rectangles are small, the smoothing can be omitted in many cases.

The resulting regions are collections of rectangles of the chosen size Row * Column. Only regions containing at least MinSize points are returned.

Regiongrowing is a very fast operation, and thus suited for time-critical applications.

Attention

Column and Row are automatically converted to odd values if necessary.

Parameters

Image (input_object)	image(-array) → object (byte / direction / cyclic / int1 / int2 / int4 / real)
Input image.

Regions (output_object)	region-array → object
Segmented regions.

Row (input_control)	extent.y → (integer)
Vertical distance between tested pixels (height of the raster).
Default value: 3
Suggested values: 1, 3, 5, 7, 9, 11, 13, 15, 17, 19, 21
Typical range of values: 1 ≤ Row ≤ 99 (lin)
Minimum increment: 2
Recommended increment: 2
Restriction: (Row >= 1) && odd(Row)

Column (input_control)	extent.x → (integer)
Horizontal distance between tested pixels (height of the raster).
Default value: 3
Suggested values: 1, 3, 5, 7, 9, 11, 13, 15, 17, 19, 21
Typical range of values: 1 ≤ Column ≤ 99 (lin)
Minimum increment: 2
Recommended increment: 2
Restriction: (Column >= 1) && odd(Column)

Tolerance (input_control)	number → (real / integer)
Points with a gray value difference less then or equal to tolerance are accumulated into the same object.
Default value: 6.0
Suggested values: 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 12.0, 14.0, 18.0, 25.0
Typical range of values: 1.0 ≤ Tolerance ≤ 127.0 (lin)
Minimum increment: 0.01
Recommended increment: 1.0
Restriction: (0 <= Tolerance) && (Tolerance < 127)

MinSize (input_control)	integer → (integer)
Minimum size of the output regions.
Default value: 100
Suggested values: 1, 5, 10, 20, 50, 100, 200, 500, 1000
Typical range of values: 1 ≤ MinSize
Minimum increment: 1
Recommended increment: 5
Restriction: MinSize >= 1

Example

read_image(Image,'fabrik')
mean_image(Image,Mean,Row,Column)
regiongrowing(Mean,Result,Row,Column,6.0,100)

Complexity

Let N be the number of found regions and M the number of points in one of these regions. Then the runtime complexity is O(N * log(M) * M).

Result

regiongrowing returns 2 (H_MSG_TRUE) if all parameters are correct. The behavior with respect to the input images and output regions can be determined by setting the values of the flags 'no_object_result', 'empty_region_result', and 'store_empty_region' with set_system. If necessary, an exception is raised.

Parallelization Information

regiongrowing is reentrant and automatically parallelized (on tuple level).

Possible Predecessors

binomial_filter, mean_image, gauss_image, smooth_image, median_image, anisotropic_diffusion

Possible Successors

select_shape, reduce_domain, select_gray

Alternatives

regiongrowing_n, regiongrowing_mean, label_to_region

Module

Foundation