TFeatureMap Struct Reference

Summarizes a number of parameters controlling feature extraction from images. More...

Data Fields
char	Code [16]
double	CorrectX
double	CorrectY
cvbint32_t	FeatureResolution
cvbint32_t	FWB
	Bottom edge of the feature window relative to the pattern position.
cvbint32_t	FWL
	Left edge of the feature window relative to the pattern position.
cvbint32_t	FWR
	Right edge of the feature window relative to the pattern position.
cvbint32_t	FWT
	Top edge of the feature window relative to the pattern position.
cvbint32_t	Interpolate
TMapMode	Mode
cvbint32_t	Normalization
cvbint32_t	NumImgPlanes
unsigned char	Reserved [128]
	Reserved for future use.
cvbint32_t	RetinaH
cvbint32_t	RetinaW

Detailed Description

Summarizes a number of parameters controlling feature extraction from images.

Similar to the TTrainParams structure, the values of the TFeatureMap structure are defined at training time, i.e. when the classifier is generated, through the parameters passed to the function that generates the classifier (e.g. PMTrainClassifierFromSil, PMTrainClassifierFromMts, PMTrainSearchClassifierFromSil or PMTrainSearchClassifierFromMts). Therefore the TFeatureMap can only be extracted for an existing classifier (PMGetClfTrainParams or PMGetSearchClfTrainParams).

Field Documentation

Code

char Code[16]

Contains the zero-terminated preprocessing string which is composed using the following characters:

• p = 4x4 Gauss pyramid (binomial low pass + sub sampling).
• a = multi-directional derivative followed by Gauss pyramid.
• s = same as a but with a simpler and slightly faster gradient filter.
• + = concatenates feature vectors, example: aa+a is feature vector generated by the preprocessing code aa concatenated with that generated by aaa.

CorrectX

double CorrectX

Horizontal retina correction factor: CorrectX := (FWR-FWL+1)/RetinaW (initialized automatically during training).

CorrectY

double CorrectY

Vertical retina correction factor: CorrectY := (FWB-FWT+1)/RetinaH (initialized automatically during training).

FeatureResolution

cvbint32_t FeatureResolution

Determines the dimensions of the retina (see feature-map lesson in the tutorial on search functions).

Interpolate

cvbint32_t Interpolate

Image interpolation mode for feature extraction. False implies simple truncation of fractional parts, True (recommended) implies bilinear interpolation.

Mode

TMapMode Mode

Type of the feature map. MM_MantoStandard currently is the only available map mode.

Normalization

cvbint32_t Normalization

Normalization constant for feature vectors. Currently a default value of 150 will always be used during training.

NumImgPlanes

cvbint32_t NumImgPlanes

Number of color planes in the images from which the classifier was trained. If this differs from the number of color planes in the image on which the classifier is used an error will be returned.

RetinaH

cvbint32_t RetinaH

Actual height of Retina in pixels (initialized automatically during training).

RetinaW

cvbint32_t RetinaW

Actual width of Retina in pixels (initialized automatically during training).