OCV & OCR - A Validation Perspective

A short and concise presentation that I shared with my team for validation approach towards OCV/OCR Inspection Systems.
This presentation shall be helpful to those who are working on Machine Vision Systems.

1. OCV & OCR A Validation Perspective

2. TABLE OF CONTENTS INTRODUCTION TO OCV/OCR APPLICATIONS OF OCV/OCR UNDERSTANDING YOUR TEST ENVIRONMENT VALIDATION APPROACH TOWARDS OCV/OCR SYSTEMS

3. INTRODUCTION What is OCV & OCR? →It is conversion of printed, documented or handwritten text into machine encoded/readable form. →It is a common method of digitizing printed texts so that they can be electronically edited, searched, stored more compactly, displayed on-line, and used in machine processes such as cognitive computing , machine translation , (extracted) text-to-speech , & text mining .

4. COMPARISON OF OCV & OCR OCV : Optical Character Verification → There is a prior knowledge about what the specific alphanumeric pattern is and the system is essentially verifying or validating that the current character is exactly what it is supposed to be → The output of an OCV is quality report of correctness. OCR: Optical Character Recognition → An alphanumeric character is being recognized based on a comparison to essentially a database of character patterns. → The output of an OCR is machine usable text. OCV: verify quality OCR: read text

5. COMPARISON OF OCV & OCR Verification: →Inspecting characters for content, correctness, quality, contrast and sharpness compared to stored templates. Examples: Date / Lot verification Component ID verification Label, carton Verification of on-line printing Clinical labels Blister packs Direct printing on product ABC-123 Use OCV to check levels of print quality and legibility

6. Reading: →A tool for reading text strings of random content and converting to machine usable text. Examples: Sorting and identification Serialization Verification of readability decoded=readable Verification of on-line printing Codes with inconsistent character placement or size Can verify text using match OCR can be used to read serial numbers on a data plate COMPARISON OF OCV & OCR

7. CLASSIFICATION OF OCV/OCR SYSTEMS →NORMAL - Normal OCR/OCV refers to detection of standard computer generated/printed characters. →ADVANCE/INTELLIGENT - I ntelligent system refers to detection of handwritten cursive text.

8. APPLICATIONS OF OCV/OCR Data entry for business documents Automatic number plate recognition Pen computing Assistive technology for blind/visually impaired users Track & trace of goods throughout the supply chain

9. UNDERSTANDING TEST ENVIRONMENT Font type FOV Vision sensor(camera) Illumination(optics) Image Focus Aspect ratio Line Weight/Stroke Width

10. FONT TYPES →Fixed Fonts : A font whose letters and characters each occupy the same amount of horizontal space. This font characters must conform to a fixed pattern. Examples: OCR-A, OCR-B: Many printed applications such as passports, documents, and pharmaceutical labels SEMI: Used for semiconductor manufacturing MICR: Banking documents such as checks. →Trainable Font: Any font can be presented and learned by Machine Vision software during setup, then identified during run-time. Examples: DOTTED UNDERSTANDING TEST ENVIRONMENT When using trainable font tools, a character is not recognized until it is trained

11. Why font type matters? → Fixed fonts are designed to ensure precisely specified sizing, spacing, and distinctive character shapes which improves accuracy, reliability & reduces processing time. → Incase of other fonts character width would be varying. UNDERSTANDING TEST ENVIRONMENT

12. Fixed Fonts provide →Uniform character spacing →Each character designed to be different than all others to maximize difference in similar characters . Example: Consider character O which has many similar characters such as Q,C & G. OCR-A ARIAL UNDERSTANDING TEST ENVIRONMENT

13. Field of View → It is the area in the eyepiece of an optical instrument in which the image is visible. →FOV will directly influence maximum no of characters that can be detected/decoded under the given camera & lens. Large FOV → Large data Small FOV → Small data UNDERSTANDING TEST ENVIRONMENT

14. Vision Sensor(Camera) →For a given FOV, camera resolution directly influences resolution i.e. character width & height in terms of pixels. →Higher the camera resolution, higher the no of pixels assigned to characters. Higher no of pixels further increases performance. UNDERSTANDING TEST ENVIRONMENT 25 Pixels 30 Pixels

15. Illumination →Illumination distinguishes foreground from background. →Illumination should be uniform throughout the FOV & strong enough such that higher contrast is generated between foreground & background. UNDERSTANDING TEST ENVIRONMENT

16. Image Focus →Before running the system user must ensure that image is focused such that character edges are sharp. UNDERSTANDING TEST ENVIRONMENT

17. Aspect Ratio →It is a ratio of width to the height of a character. →It does not influence character decoding. It is a pre-condition i.e. technical specification of OCV/OCR systems specify minimum character width, height system can decode. UNDERSTANDING TEST ENVIRONMENT A Height Width

18. Line Weight/Stroke Width →The weight of a particular font is the thickness of the character outlines relative to their height. →As a rule of thumb, stroke width of the smallest character should be at least three pixels wide. →Specification of an OCV/OCR system should specify minimum width system will be able to decode. It is specified in terms of pixels. UNDERSTANDING TEST ENVIRONMENT A Line Weight

19. →Apart from functional, OCV/OCR systems should be validated focusing more on non-functional points. →Considering area of application an OCV/OCR system should be validated for 1. Sensitivity and Specificity 2. Usability 3. Accuracy 4. Flexibility(Pattern Matching) VALIDATION APPROACH

20. Teach & Inspection →Once all characters are taught system must apply teach data correctly in inspection. →As long as the printing specification remains same as taught characters, characters should be decoded correctly. VALIDATION APPROACH - FUNCTIONAL

21. Decode string & status →For each image inspected system must reflect the decoded string along with status. →If string is decoded correctly it should be reflected in GREEN or else if it is failed then it should be reflected in RED . VALIDATION APPROACH - FUNCTIONAL

22. Verification score →For each character read or measured by system , a verification score is assigned to the character. →In Inspection system compares the measured score of each character to the reference score set in teach. →If measured score < reference score character should be Failed & if measured score > reference score character should be Passed . →System should allow user to set value of reference score for each character in Teach as well as Inspection. VALIDATION APPROACH - FUNCTIONAL

23. Consistency & Reliability → An OCV/OCR system must produce similar results under consistent environment. → Environment comprises of optics(light & polarizer) & vision sensors(camera & lens). Test Method : Keeping carton steady at a position under the camera, tester can generate triggers & observe character decoding for each trigger. Trigger count can be kept at 100. VALIDATION APPROACH – NON FUNCTIONAL

24. Usability →An OCV/OCR system should be usable enough to gain confidence of the end user. →Following points must be considered by a tester from usability perspective. 1. System should be very easily programmable 2. System should be assistive, preventing user from false teach VALIDATION APPROACH - NON FUNCTIONAL

25. Accuracy & Precision →An OCV/OCR system is considered valid if it is both accurate & precise. A tester must validate the system for accuracy & precision. →As per the industry standards OCV/OCR systems are considered best when accuracy of 99.9% is given. →System should be robust enough to handle confusion pairs irrespective if any font type such as 'B' with a '3' or '8'; letter 'I' with a '1'; 'G' with '6'; the letter 'O' with the numeral zero; '6' with '9' or '3' with'8‘. VALIDATION APPROACH – NON FUNCTIONAL

26. Flexibility (Pattern Matching) →System should not be restrictive. It should be flexible enough to accommodate possible variations. →These variations can be in terms of print orientation & product alignment along vertical & horizontal direction. →Print orientation variation can be handled through functionality of image rotation. →Product alignment variation can be handled through pattern matching. VALIDATION APPROACH – NON FUNCTIONAL

27. VARIATIONS DURING INSPECTION & WAYS TO TACKLE THEM Carton alignment variation Print alignment variation Print quality variation

28. Carton alignment variation →During inspection position of the carton may vary along horizontal or vertical direction with respect to its original teach position. →These positional variations can lead to false rejection in inspection. →System shall be able to tackle such a situation through Pattern Matching option. →Pattern matching enables the system to read character string at any position within FOV. VARIATIONS DURING INSPECTION & WAYS TO TACKLE THEM

29. Print alignment variations →There may be products with up side down printing. Position of the camera with respect to printed characters may not be aligned. →Through Image Rotation functionality such a scenario can be handled. →Provision should be there within system to teach & inspect images with rotation of 90,180,270 degrees. VARIATIONS DURING INSPECTION & WAYS TO TACKLE THEM

30. Print quality variations →In case of OCV/OCR system along with on line printing, quality of print may deteriorate after certain time period. →Variation in print quality also leads to false rejection. →Such variations can be handled by reducing verification score during live inspection. →Integrity & readability of the data must be ensured even if verification score is reduced. VARIATIONS DURING INSPECTION & WAYS TO TACKLE THEM Dot Size Overprint Underprint

31. THANK YOU