当前位置:文档之家› IS07_1_OCVOCR

IS07_1_OCVOCR

Optical Character Verification & Recognition

Objectives
? The student will correctly identify:
? the maximum number of instances possible for an OCR character model ? the basic purpose of tuning for OCR/OCV ? the conditions necessary for strings to pass OCV or OCR
? The student will correctly describe & apply to a sample image Property Sheet parameters & auto-inserted information for the three OCV/OCR functions TrainFont, VerifyText, and ReadText including
? Tuning ? Field Definitions ? Field Strings
2

Improved OCV/OCR Tool
? Font Training Improvements
? Easier Font Training ? Higher confidence character reading & verification
? Run-time Algorithm Improvements
? Accuracy Mode
? For difficult to read strings
? Scale & Confusing Background Tolerance
? Better handles runtime character variation
Access to tools remains unchanged in In-Sight Explorer interface
3

OCV / OCR: Overview
? Both require a database of trained models of characters
TrainFont function
? OCV : you specify a desired string and search for match
VerifyText function
? OCR: determines each character in unknown string
ReadText function
4

Creating a Font: TrainFont
? Trained font = up to 64 character
models
? Each character model based on 1-8 trained instances of that character ? Can have multiple fonts, each defined by different TrainFont ? Best to use special fonts designed for OCR/OCV
5

Font Training Dialog
? Font Name - Optional for: multiple font libraries ? String – ensure matching case, add spaces if training string has spaces
6

Font Training Dialog
1. 2.
Select Region Select Segment Verify each character has a box and corresponding identifier, if not, use the Segment Adjustments Select Add to Font to create a MyFont folder
7
3.

Adding Characters
Redrawing the region
Extending the region over existing characters - select Train New
Redrawing the region on a single character – not recommended
8

Font Training Dialog
? Select the Font Tab ? Verify characters are correct, or go back to the training tab to add characters ? Select OK Note: up to this point, the font folder has not been saved until you click on OK
9

Font Editor
? User can edit the binary model of the character ? Improves appearance of poorly trained characters ? Removes noise and unwanted features ? Very useful when:
? ? ? Number and quality of training images is limited Characters are not well formed Background is noisy
10

Multiple Instances
? Maximum of 8 instances per character model ? Additional instances increase variation tolerance ? If you don’t want to tolerate much variation, train on a single good example ? If you want to tolerate variation, train on multiple instances that show the variation ? Do not train on different size fonts make a separate library for each size
11

Instances
? Ensure character print is as consistent as possible ? Dramatic changes in shape & size create poor models & fail during run-time
Consistent characters create good model
Inconsistent characters create confusing model
12

相关主题
文本预览
相关文档 最新文档