create_ocr_class_knnT_create_ocr_class_knnCreateOcrClassKnnCreateOcrClassKnn (Operator)

Name

create_ocr_class_knnT_create_ocr_class_knnCreateOcrClassKnnCreateOcrClassKnn — Create an OCR classifier using a k-Nearest Neighbor (k-NN) classifier.

Signature

Description

create_ocr_class_knncreate_ocr_class_knnCreateOcrClassKnnCreateOcrClassKnnCreateOcrClassKnn creates an OCR classifier that uses a k-Nearest Neighbor (k-NN). The handle of the k-NN classifier is returned in OCRHandleOCRHandleOCRHandleOCRHandleOCRHandle.

For a description on how a k-NN works, see create_class_knncreate_class_knnCreateClassKnnCreateClassKnnCreateClassKnn.

The length of the feature vector of the k-NN is determined from the features that are used for the OCR, which are passed in FeaturesFeaturesFeaturesFeaturesfeatures. The features are described below. The number of classes is determined from the names of the characters which are passed in CharactersCharactersCharactersCharacterscharacters.

FeaturesFeaturesFeaturesFeaturesfeatures can contain a tuple of several feature names. Each of these names results in one or more features to be calculated for the classifier. Some of the feature names compute gray value features (e.g., 'pixel_invar'"pixel_invar""pixel_invar""pixel_invar""pixel_invar"). Because a classifier requires a constant number of features (input variables), a character to be classified is transformed to a standard size, which is determined by WidthCharacterWidthCharacterWidthCharacterWidthCharacterwidthCharacter and HeightCharacterHeightCharacterHeightCharacterHeightCharacterheightCharacter. The interpolation to be used for the transformation is determined by InterpolationInterpolationInterpolationInterpolationinterpolation. It has the same meaning as in affine_trans_imageaffine_trans_imageAffineTransImageAffineTransImageAffineTransImage. The interpolation should be chosen such that no aliasing effects occur in the transformation. For most applications, InterpolationInterpolationInterpolationInterpolationinterpolation = 'constant'"constant""constant""constant""constant" should be used. It should be noted that the size of the transformed character is not chosen too large, because the generalization properties of the classifier may become bad for large sizes. In particular, large sizes will cause small segmentation errors to have a large influence on the computed features if gray value features are used. This happens because segmentation errors will change the smallest enclosing rectangle of the regions, which results in characters are zoomed differently than the characters in the training set. In most applications, sizes between 6x8 and 10x14 should be used.

The parameter FeaturesFeaturesFeaturesFeaturesfeatures can contain the following feature names for the classification of the characters.

'default'"default""default""default""default":

'ratio'"ratio""ratio""ratio""ratio" and 'pixel_invar'"pixel_invar""pixel_invar""pixel_invar""pixel_invar" are selected.

'pixel:'"pixel:""pixel:""pixel:""pixel:"

Gray values of the character (WidthCharacterWidthCharacterWidthCharacterWidthCharacterwidthCharacter x HeightCharacterHeightCharacterHeightCharacterHeightCharacterheightCharacter features).

'pixel_invar:'"pixel_invar:""pixel_invar:""pixel_invar:""pixel_invar:"

Gray values of the character with maximum scaling of the gray values (WidthCharacterWidthCharacterWidthCharacterWidthCharacterwidthCharacter x HeightCharacterHeightCharacterHeightCharacterHeightCharacterheightCharacter features).

'pixel_binary:'"pixel_binary:""pixel_binary:""pixel_binary:""pixel_binary:"

Region of the character as a binary image zoomed to a size of WidthCharacterWidthCharacterWidthCharacterWidthCharacterwidthCharacter x HeightCharacterHeightCharacterHeightCharacterHeightCharacterheightCharacter (WidthCharacterWidthCharacterWidthCharacterWidthCharacterwidthCharacter x HeightCharacterHeightCharacterHeightCharacterHeightCharacterheightCharacter features).

'gradient_8dir:'"gradient_8dir:""gradient_8dir:""gradient_8dir:""gradient_8dir:"

Gradients are computed on the character image. The gradient directions are discretized into 8 directions. The amplitude image is decomposed into 8 channels according to these discretized directions. 25 samples on a 5x5 grid are extracted from each channel. These samples are used as features (200 features).

'projection_horizontal:'"projection_horizontal:""projection_horizontal:""projection_horizontal:""projection_horizontal:"

Horizontal projection of the gray values (see gray_projectionsgray_projectionsGrayProjectionsGrayProjectionsGrayProjections, HeightCharacterHeightCharacterHeightCharacterHeightCharacterheightCharacter features).

'projection_horizontal_invar:'"projection_horizontal_invar:""projection_horizontal_invar:""projection_horizontal_invar:""projection_horizontal_invar:"

Maximally scaled horizontal projection of the gray values (HeightCharacterHeightCharacterHeightCharacterHeightCharacterheightCharacter features).

'projection_vertical:'"projection_vertical:""projection_vertical:""projection_vertical:""projection_vertical:"

Vertical projection of the gray values (see gray_projectionsgray_projectionsGrayProjectionsGrayProjectionsGrayProjections, WidthCharacterWidthCharacterWidthCharacterWidthCharacterwidthCharacter features).

'projection_vertical_invar:'"projection_vertical_invar:""projection_vertical_invar:""projection_vertical_invar:""projection_vertical_invar:"

Maximally scaled vertical projection of the gray values (WidthCharacterWidthCharacterWidthCharacterWidthCharacterwidthCharacter features).

'ratio:'"ratio:""ratio:""ratio:""ratio:"

Aspect ratio of the character (see height_width_ratioheight_width_ratioHeightWidthRatioHeightWidthRatioHeightWidthRatio, 1 feature).

'anisometry:'"anisometry:""anisometry:""anisometry:""anisometry:"

Anisometry of the character (see eccentricityeccentricityEccentricityEccentricityEccentricity, 1 feature).

'width:'"width:""width:""width:""width:"

Width of the character before scaling the character to the standard size (not scale-invariant, see height_width_ratioheight_width_ratioHeightWidthRatioHeightWidthRatioHeightWidthRatio, 1 feature).

'height:'"height:""height:""height:""height:"

Height of the character before scaling the character to the standard size (not scale-invariant, see height_width_ratioheight_width_ratioHeightWidthRatioHeightWidthRatioHeightWidthRatio, 1 feature).

'zoom_factor:'"zoom_factor:""zoom_factor:""zoom_factor:""zoom_factor:"

Difference in size between the character and the values WidthCharacterWidthCharacterWidthCharacterWidthCharacterwidthCharacter and HeightCharacterHeightCharacterHeightCharacterHeightCharacterheightCharacter (not scale-invariant, 1 feature).

'foreground:'"foreground:""foreground:""foreground:""foreground:"

Fraction of pixels in the foreground (1 feature).

'foreground_grid_9:'"foreground_grid_9:""foreground_grid_9:""foreground_grid_9:""foreground_grid_9:"

Fraction of pixels in the foreground in a 3x3 grid within the smallest enclosing rectangle of the character (9 features).

'foreground_grid_16:'"foreground_grid_16:""foreground_grid_16:""foreground_grid_16:""foreground_grid_16:"

Fraction of pixels in the foreground in a 4x4 grid within the smallest enclosing rectangle of the character (16 features).

'compactness:'"compactness:""compactness:""compactness:""compactness:"

Compactness of the character (see compactnesscompactnessCompactnessCompactnessCompactness, 1 feature).

'convexity:'"convexity:""convexity:""convexity:""convexity:"

Convexity of the character (see convexityconvexityConvexityConvexityConvexity, 1 feature).

'moments_region_2nd_invar:'"moments_region_2nd_invar:""moments_region_2nd_invar:""moments_region_2nd_invar:""moments_region_2nd_invar:"

Normalized 2nd moments of the character (see moments_region_2nd_invarmoments_region_2nd_invarMomentsRegion2ndInvarMomentsRegion2ndInvarMomentsRegion2ndInvar, 3 features).

'moments_region_2nd_rel_invar:'"moments_region_2nd_rel_invar:""moments_region_2nd_rel_invar:""moments_region_2nd_rel_invar:""moments_region_2nd_rel_invar:"

Normalized 2nd relative moments of the character (see moments_region_2nd_rel_invarmoments_region_2nd_rel_invarMomentsRegion2ndRelInvarMomentsRegion2ndRelInvarMomentsRegion2ndRelInvar, 2 features).

'moments_region_3rd_invar:'"moments_region_3rd_invar:""moments_region_3rd_invar:""moments_region_3rd_invar:""moments_region_3rd_invar:"

Normalized 3rd moments of the character (see moments_region_3rd_invarmoments_region_3rd_invarMomentsRegion3rdInvarMomentsRegion3rdInvarMomentsRegion3rdInvar, 4 features).

'moments_central:'"moments_central:""moments_central:""moments_central:""moments_central:"

Normalized central moments of the character (see moments_region_centralmoments_region_centralMomentsRegionCentralMomentsRegionCentralMomentsRegionCentral, 4 features).

'moments_gray_plane:'"moments_gray_plane:""moments_gray_plane:""moments_gray_plane:""moments_gray_plane:"

Normalized gray value moments and the angle of the gray value plane (see moments_gray_planemoments_gray_planeMomentsGrayPlaneMomentsGrayPlaneMomentsGrayPlane, 4 features).

'phi:'"phi:""phi:""phi:""phi:"

Sinus and cosinus of the orientation (angle) of the character (see elliptic_axiselliptic_axisEllipticAxisEllipticAxisEllipticAxis, 2 feature).

'num_connect:'"num_connect:""num_connect:""num_connect:""num_connect:"

Number of connected components (see connect_and_holesconnect_and_holesConnectAndHolesConnectAndHolesConnectAndHoles, 1 feature).

'num_holes:'"num_holes:""num_holes:""num_holes:""num_holes:"

Number of holes (see connect_and_holesconnect_and_holesConnectAndHolesConnectAndHolesConnectAndHoles, 1 feature).

'cooc:'"cooc:""cooc:""cooc:""cooc:"

Values of the binary cooccurrence matrix (see gen_cooc_matrixgen_cooc_matrixGenCoocMatrixGenCoocMatrixGenCoocMatrix, 8 features).

'num_runs:'"num_runs:""num_runs:""num_runs:""num_runs:"

Number of runs in the region normalized by the height (1 feature).

'chord_histo:'"chord_histo:""chord_histo:""chord_histo:""chord_histo:"

Frequency of the runs per row (not scale-invariant, HeightCharacterHeightCharacterHeightCharacterHeightCharacterheightCharacter features).

After the classifier has been created, it is trained using trainf_ocr_class_knntrainf_ocr_class_knnTrainfOcrClassKnnTrainfOcrClassKnnTrainfOcrClassKnn. After this, the classifier can be saved using write_ocr_class_knnwrite_ocr_class_knnWriteOcrClassKnnWriteOcrClassKnnWriteOcrClassKnn. Alternatively, the classifier can be used immediately after training to classify characters using do_ocr_single_class_knndo_ocr_single_class_knnDoOcrSingleClassKnnDoOcrSingleClassKnnDoOcrSingleClassKnn or do_ocr_multi_class_knndo_ocr_multi_class_knnDoOcrMultiClassKnnDoOcrMultiClassKnnDoOcrMultiClassKnn.

A comparison of the k-NN and the support vector machine (SVM) (see create_ocr_class_svmcreate_ocr_class_svmCreateOcrClassSvmCreateOcrClassSvmCreateOcrClassSvm) typically shows that SVMs are generally slower at training, especially for huge training sets, but achieve slightly better recognition rates than k-NNs. Please note that this guideline assumes optimal tuning of the parameters of the SVM.

Execution Information

• Multithreading type: reentrant (runs in parallel with non-exclusive operators).
• Multithreading scope: global (may be called from any thread).
• Processed without parallelization.

This operator returns a handle. Note that the state of an instance of this handle type may be changed by specific operators even though the handle is used as an input parameter by those operators.

Parameters

Example (HDevelop)

read_image (Image, 'letters')
* Segment the image.
binary_threshold(Image,&Region, 'otsu', 'dark', &UsedThreshold);
dilation_circle (Region, RegionDilation, 3.5)
connection (RegionDilation, ConnectedRegions)
intersection (ConnectedRegions, Region, RegionIntersection)
sort_region (RegionIntersection, Characters, 'character', 'true', 'row')
* Generate the training file.
count_obj (Characters, Number)
Classes := []
for J := 0 to 25 by 1
    Classes := [Classes,gen_tuple_const(20,chr(ord('a')+J))]
endfor
Classes := [Classes,gen_tuple_const(20,'.')]
write_ocr_trainf (Characters, Image, Classes, 'letters.trf')
* Generate and train the classifier.
read_ocr_trainf_names ('letters.trf', CharacterNames, CharacterCount)
create_ocr_class_knn (8, 10, 'constant', 'default', CharacterNames, \
                      [], [], OCRHandle)
trainf_ocr_class_knn (OCRHandle, 'letters.trf', [], [])
* Re-classify the characters in the image.
do_ocr_multi_class_knn (Characters, Image, OCRHandle, Class, Confidence)

Result

If the parameters are valid, the operator create_ocr_class_knncreate_ocr_class_knnCreateOcrClassKnnCreateOcrClassKnnCreateOcrClassKnn returns the value 2 (H_MSG_TRUE). If necessary, an exception is raised.

Possible Successors

trainf_ocr_class_knntrainf_ocr_class_knnTrainfOcrClassKnnTrainfOcrClassKnnTrainfOcrClassKnn

Alternatives

create_ocr_class_svmcreate_ocr_class_svmCreateOcrClassSvmCreateOcrClassSvmCreateOcrClassSvm

See also

do_ocr_single_class_knndo_ocr_single_class_knnDoOcrSingleClassKnnDoOcrSingleClassKnnDoOcrSingleClassKnn, do_ocr_multi_class_knndo_ocr_multi_class_knnDoOcrMultiClassKnnDoOcrMultiClassKnnDoOcrMultiClassKnn, clear_class_knnclear_class_knnClearClassKnnClearClassKnnClearClassKnn, create_class_knncreate_class_knnCreateClassKnnCreateClassKnnCreateClassKnn, trainf_ocr_class_knntrainf_ocr_class_knnTrainfOcrClassKnnTrainfOcrClassKnnTrainfOcrClassKnn, classify_class_knnclassify_class_knnClassifyClassKnnClassifyClassKnnClassifyClassKnn

Module

OCR/OCV