文稿康耐視insight學(xué)習(xí)ocv

1、Optical Character VerificationObjectivesThe student will correctly:Create and configure an OCV FontConfigure an OCV Pattern layoutSpecify a reasonable OCV pattern positionAnalyze the results returned from an OCV toolOCVOptical Character Verification is used to verify that a given character string is

2、 presentCommonly used to verify lot and date codesReturns TRUE if all characters in the string are correctly identified; FALSE if notUser trains the font for the characters in the string to verifyCan also be used as a “l(fā)azy” Optical Character Recognition tool ExampleVerify the lot number “04149”Gene

3、ral Steps1. Choose a Font1A. Edit a font1B. Load an existing font2. Configure the String Pattern3. Define the Pattern Position4. Choose Graphics to Display5. Test and verify resultsAdd OCV ToolAdd an OCV tool in QuickStart and link an image to its Input ImageIn this example, we will assume (for now)

4、 that the code is approximately in the same place in each imageNo fixturing neededFont TrainingFontTo start a new font, open the Font EditorAdd Normal Font ModelsChoose Add Normal Font Models to add characters to the fontIt launches the “Add Font Models” dialogFont Reference ImageIf you havent alrea

5、dy passed an image to the OCV tool, press “Grab Reference Image” to get an image that contains the characters you want to add to the fontChoose a Region ModeExtract Model CandidatesEnter the Candidates to ExtractCan be a single character or several at the same timeA box for each candidate will appea

6、r on the reference imageCharacter RegionsPosition each box around a single characterIf there is overlap, well take care of it with masking laterFont GuidelinesModels should be representative of typical run-time charactersSize is ideally between 10 x 10 pixels and 64 x 64 pixelsStroke width should no

7、t vary by more than 10%No more than 10% of the character should be missingFor dot-matrix, use the Image Processing tool to fill in the character stroke for both font training and run-time imagesThe font model image should contain a border of white space around the character 1 - 3 times the stroke wi

8、dthOrigin should be the same for all characters in a font i.e. define the origin for characters to be the center of the character baseline (For more details refer to the documentation)Extract CandidatesSelect Extract Model CandidatesA thumbnail of each candidate appears in the dialogAdd Models to Fo

9、ntBlank Font ModelsIf needed, add blank font models using a procedure similar to that for adding normal font modelsModel Masking and EditingUse Model Image Editor to edit the image to make a “better” modelEspecially useful when no “ideal” character exists for you train fromUse Mask Editor to mask ou

10、t features you do not want to train in the modelCompile the FontCompiling a font creates a Confusion MatrixIndications of models that have high correlations and may be confused at run-timeYou must compile a font to verify a patternConfusion MatrixIndicates on a scale of 0 1.0 how much confusion exis

11、ts between two characters in a fontOnly those characters whose confusion number exceeds the threshold will be comparedTraining Multiple Models of a CharacterIn general, train enough instances of a character to capture the expected variationsHowever, the more instances, the slower verification can be

12、Well see our options available for verifying in the Pattern set-upPattern LayoutPattern CharactersGrab a Reference ImageGo to Edit and Layout for Simple Pattern Set-upSimple Pattern Set-upIndicate the string to be verifiedSelect a baseline direction Where to look for the string in the imageLocate Al

13、l will find all characters in the area specifiedAdvanced Pattern Set-upCharacter UncertaintySet uncertainty values for the charactersUncertainty from reference image positionsGreater uncertainty = Greater execution timeThresholdsAccept is the minimum correlation score a character must receive to be

14、considered a valid matchConfidence is the amount that a characters matching score must exceed the highest scoring confusing character in order to pass verificationPattern UncertaintiesPattern Uncertainties are the expected variation of the overall pattern from the pattern originPattern origin so far

15、 has assumed to be the image originIncreasing uncertainty will increase execution timeRunning OCVOCV ResultsPattern StatusOverall whether the pattern was correctly verified or notPattern ScoreThe average correlation of the characters in the run-time string to the trained patternOCV ResultsCharacter

16、results return values for each character in the stringPatterns That MoveExpected PoseIf you pattern is going to move significantly from image to image, you will probably want to supply an Expected PoseMay come from either fixturing toolOr from a PMAlign tool (our example)Expected PoseIn this example, find the “Lot” pattern for the expected posePattern OriginAfter linking the appropriate terminals, the additional step is to duplicate the origin result of the PMAlign to t