PUBLIC

ChromaThresholdRGBC::ChromaThresholdRGBC(void) ChromaThresholdRGBC::ChromaThresholdRGBC(RealT,RealT,RealT,RealT,RealT,RealT,RealT,ByteT,ByteT,ByteT) ChromaThresholdRGBC::ChromaThresholdRGBC(const ImageC &,RealT,RealT,ByteT,ByteT,ByteT) ChromaThresholdRGBC::Apply(ImageC &,const ImageC &) const ChromaThresholdRGBC::Apply(ImageC &,const ImageC &) const ChromaThresholdRGBC::ApplyCopyAlpha(ImageC &,const ImageC &,ImageC &) const ChromaThresholdRGBC::ApplyCopyAlpha(ImageC &,ImageC &)

This is a multi-purpose class for searching for a known colour in an rgb image. Can be used for skin colour detection, blue-screening, etc. Is intended as an easy way to try out new ideas, rather than a very efficient or accurate method.

RGB is normalised by intensity to be robust to illumination changes. The range of colours to be found are modelled as a gaussian. The (incorrect) assumption is made that normalised rgb components are independent, but this seems to work well for many applications. The gaussian can either be constructed directly, or computed from an example image containing only the required colour.

ChromaThresholdRGBC()

default constructor

ChromaThresholdRGBC(RealT nr0,RealT ng0,RealT nb0,RealT nrw = 33.33,RealT ngw = 33.33,RealT nbw = 33.33,RealT nblack_thresh = 0.1,ByteT nlabel_match = 255,ByteT nlabel_no_match = 0,ByteT nlabel_black = 0)

Construct with user-specified params of the gaussian

nr0	mean red component
ng0	mean green component
nb0	mean blue component
nrw	inverse variance of red component
ngw	inverse variance of green component
nbw	inverse variance of blue component
nblack_thresh	luminance in range (0,1] below which a pixel is considered black rather than coloured.
nlabel_match	label to give to pixels which match the specified colour
nlabel_no_match	label to give to pixels which do not match the specified colour
nlabel_black	label to give to black pixels (ones below nblack_thresh)

Should have nr0 + ng0 + nb0 = 1 for a valid model.

ChromaThresholdRGBC(const ImageC<ByteRGBValueC> & image,RealT tolerance = 1.0,RealT black_thresh = 0.01,ByteT nlabel_match = 255,ByteT nlabel_no_match = 0,ByteT nlabel_black = 0)

Construct from example image

image	example image containing only the required colour
nblack_thresh	luminance in range (0,1] below which a pixel is considered black rather than coloured.
nlabel_match	label to give to pixels which match the specified colour
nlabel_no_match	label to give to pixels which do not match the specified colour
nlabel_black	label to give to black pixels (ones below nblack_thresh)

void Apply(ImageC<ByteT> & result,const ImageC<ByteRGBValueC> & image) const

perform threshold on RGB image and return binary result

void Apply(ImageC<ByteT> & result,const ImageC<ByteRGBAValueC> & image) const

perform threshold on RGB image and return binary result

void ApplyCopyAlpha(ImageC<ByteT> & result,const ImageC<ByteRGBAValueC> & image,ImageC<ByteRGBAValueC> & auximage) const

perform threshold on RGBA image, return binary result and copy results to alpha channel of aux image

void ApplyCopyAlpha(ImageC<ByteT> & result,ImageC<ByteRGBAValueC> & image)

perform threshold on RGBA image, return binary result and copy results to alpha channel of original image