PUBLIC

RegOrderedGrowC::RegOrderedGrowC(UIntT) RegOrderedGrowC::Grow(const ByteImageC &,const DListC &) const RegOrderedGrowC::Label(const ImageRectangleC &,const DListC &) const RegOrderedGrowC::Segment(const ByteImageC &,const DListC &) const RegOrderedGrowC::Overlay(const RGBImageC &,const ByteRGBValueC &,const DListC &) const

RegOrderedGrowC
Segmentation using region growing ordered by boundary pixel intensity.
include	"amma/DERA/RegOrderedGrow.hh"
User Level:	Default
Library:	DERARegTrack
Example:	exIndex.cc
Section:	Applications.Image Processing.DERA Small Target Detection
In Scope:	std

Comments:
This algorithm is a fast implementation of the segmentation algorithm by A. Hojjatoleslami. Speed is achieved by using the following implementation techniques:

• A priority queue data structure stores the pixels in the current boundary. This makes the search for the next pixel to add to the region very fast.
• A RegBoundaryC object is used to store the region. This is an AMMA big object and is therefore effectively very fast to pass into functions and return.
• The RegBoundaryC object calculates all region statistics incrementally so does not need to traverse the whole region at each iteration in order to calculate the average and gradient boundary values.
• The number of iterations that the segmentation grows for is typically left low compared to the values used by Ali since it makes segmentation a lot faster. The disadvantage is that it undersegments large objects, but they should be detected and removed by later processing stages.

UIntT _maxN;

ind	Index of types of intererest.

RegOrderedGrowC(UIntT maxN)

Constructor

ind	Index of types of intererest.

const DListC<RegBoundaryC> Grow(const ByteImageC & scene,const DListC<RegSeedC> & seedList) const

Grows a region from each seed and returns the list of boundaries

ind	Index of types of intererest.

Regions will be assigned label corresponding to the one of their seed. There is absolutely no check at this point for whether or not regions overlap.

ImageC<NumLabelC> Label(const ImageRectangleC & rect,const DListC<RegBoundaryC> & boundList) const

Generates a labelled image using the list of boundaries

ind	Index of types of intererest.

Labelled regions in the image cannot overlap and as they are going to be added in the order that they appear in the boundList, if they do, previously added regions will be overwritten. The internal boundary is highlighted by setting bit 16.

ImageC<NumLabelC> Segment(const ByteImageC & scene,const DListC<RegSeedC> & seedList) const

Grows a region from each seed and returns a labelled image

ind	Index of types of intererest.

The job is performed internally by calling first Grow and then Label.

RGBImageC Overlay(const RGBImageC & scene,const ByteRGBValueC & colour,const DListC<RegBoundaryC> & boundList) const

Generates an overlaid image of the scene with region outlines

ind	Index of types of intererest.

The internal boundary of each region is highlighed in an output overlaid image of the scene.