VisionX V4 Documentation

TEXT="#000000" >

VisionX V4 Subject Index Ordered Index General Info. Home

Feature Vector Commands

VisionX V4

The "vfeature" module is comprised of commands for the management and classification of feature vector data. Data files for feature vectors are text files using a comma-separated-values (csv) format. In addition there there are several different legacy feature vector formats available in VisionX file format that are targeted for different applications. These files may be converted to standard csv format files by using the VisionX utility commands vexcel or vxltovf.

The data format for the classifiers is a csv file with a header row; each subsequent row contains a data vector for classification. The columns contain a set of identifiers followed by a column labeled "class" which provides the class of the data set (blank if unknown), followed by a set of features. After classification the file is returned with the feature values replaced with the classifier response.

Command	Function: Classification
vrclasstt	Feature Classification for a two class problem
vrclasscv	Cross validation Classification analysis for a two class problem
Command	Function: Import
vexcel	Creates a comma separated values (csv) file from a simple VisionX feature vector format file
vxltovf -i	Creates a comma separated values (csv) file from a VisionX stuctured feature file format
Command	Function: Data Selection
vcsvop	Manipulates csv files including selecting specific columns
vciop	Manipulates csv files with "case,image" identifiers including selecting specific columns

Multi-class feature commands and files

The following commands support the native "vclasf" multi-class nearest neighbor classifier. For two class problems the use of this classifier has been depreciated in favor of the above classifier commands. VisionX feature vector data for "vclasf" classification are stored in a VisionX file with the structure listed below or in text files with a comma-separated-values (csv) format. The commands related to the vclasf system are as follows:

Command	Function: Import/Export
vxltovf	Creates a VisionX feature vector format file from a comma separated or tab separated file
Command	Function: Feature manipulation
vfvcat	Concatenates feature vectors together
vfvpick	Selects a subset of features from a feature file
Command	Function: Classification
vclasf	Performs classification of feature vectors using a nearest neighbors algorithm

Legacy File Structure

All the programs utilize the VisionX feature vector file format. Feature vectors are stored in framed VisionX files according to the following structure (for each frame):

Element	Data type	Required?
Header	VX_GFH	(required)
ID	VX_ID	(optional)
Class	VX_GSINT	(optional)
Description	VX_DESC	(optional)
Features	VX_GFV	(required)

A header is only allowed once per frame. Only one ID, class, and description are allowed for each feature vector; the programs assume that the ID and class will be present immediately preceding the feature vector in the specified order, the description is optional and is a free form description with only one constraint -- it can not contain "tab" or "," characters. Within a frame, all the feature vectors must have the same length. Output files from classification programs have the following structure (for each frame):

Element	Data type	Present?	Comments
Header	VX_GFH	always
ID	VX_ID	only if present in input
Class	VX_GSINT	only if present in input
Description	VX_DESC	only if present in input
Class vector	VX_GINT	always	First element is classified class, remaining elements are class of each of the nearest neighbors
Distance vector	VX_DIST	always	First element is distance to classified class, remaining elements are distances to each of the nearest neighbors
ID vector	VX_IDV	only if ID present in input	First element is ID of closest feature vector of classified class, remaining elements are ID of each of the nearest neighbors