HANDWRITING
RECOGNITION
Introduction
 Ability of computer to receive and interpret understandable
handwritten inputs
 Inputs include paper document, handwritten, photograph
or from input devices
 Input may be “off line” from piece of paper Offline
Handwriting Recognition or “on line” from movement of
pens Online Handwriting Recognition
 “A complete handwriting recognition system also handles
formatting, performs correct segmentation into characters
and finds the most plausible words.”
A segment of handwriting is a piece of the pen-tip trajectory
between two defined segmentation points
Offline Character Recognition
 Optical Character Recognition(OCR):
 Mechanical or electronic conversion of images typed, handwritten or printed
text into machine-encoded text
 Scanner also scans the texts documents but OCR is different as it scans each
letters and characters in the document and saves it in text document whereas
scanners scans the documents and saves it as image, not editable
Implementations of OCR
1. Optophone(1913):
 used selenium photo sensors to detect black print and
convert it into an audible output which could be
interpreted by a blind person
 Works on the special behaviour of selenium which
conducts electricity differently on light and darkness
 As it scans on page it distinguish between dark ink of
text and lighter blank spaces
 It generates tones corresponding to different letters
 E.g. Every vertical line produces a chord then m has three chords and n has two
chords
 Only a few units were built and reading was initially exceedingly slow
 Later models of the Optophone allowed speeds of up to 60 words per minute,
though only some subjects are able to achieve this rate
2. In 1931 a machine was developed to convert printed text to telegraph.
First one to translate printed text to electrical impulse
3. Optacon(1960s-1970s):
 Uses cameras to focus on the letters and replicates the letters on the device
and produces 144 vibrating reads which is sensed by index finger
OCR Techniques
1. Pre-Processing:
 De-Skew: makes lines of text perfectly horizontal or vertical
 Despeckle: removes noises from image and make smooth edges
 Binarisation: converts any image to greyscale and separates text from
background
 Layout analysis: identifies columns, paragraphs, captions
 Line and word detection
 Script recognition
 Normalise aspect ratio and scale
2. Character Recognition:
Uses two basic character recognition algorithm:
a) Matrix Matching/ Pattern Matching:
 Compares an image to store glyph on a pixel by pixel basis
 Glyph: elemental symbol within agreed set of symbols
 Works only if input glyph are separated correctly from rest of image and it
matches the stored symbol in similar font in same scale
b) Feature Extraction:
 Decomposes glyphs into features like lines, closed loops, line directions and line
intersections
 These features are compared with an abstract vector-like representation of a
character
 K-nearest algorithms are used to compare image feature with stored glyph features
and nearest match is chose
 Feature detection in computer vision: computing abstract image
information and making local decisions at every image point whether there
is an image feature of a given type at that point or not
 Software like cuneiform and tesseract uses a two-pass approach to
character recognition known as adaptive recognition
 Adaptive recognition uses the letter shape recognised with high confidence
at first pass to recognise better the remaining letters on the second pass
 It is useful for unusual fonts and low-quality scans
3. Post-processing:
 OCR accuracy can be increased if the output is compared with the letters or
words that makes meaningful words or sentence respectively but might not be
useful for proper nouns. E.g. I a$e rice can be traced to I ate rice but $am ate
rice cannot be traced to Ram ate rice as it could be Pam ate rice
 Co-occurrence frequencies can be used by noting certain words seen together.
for e.g. “Washington D.C” is seen more often than “Washington DOC”
 English grammar can be used to determine whether word is noun or verb
Accuracy
 Study based on recognition of 19th and early 20th century newspaper
concluded that character-by-character OCR accuracy for commercial OCR
varied from 81%-99%
Application
 Automatic number plate recognition
 Extracting business card information into a contact list
 Make electronic images of printed documents searchable
 Assistive technology for blind and visually impaired users
 Language translation for e.g. of road signs
 Online in the sense that inputs are written on the screen instead of printed
documents or written paper sheets
 Sensor picks-up the pen-tip movements as well as pen-up/pen-down
switching
 This kind of data is known as digital ink and can be regarded as the digital
representation of handwriting
 The sensor’s signal is converted into letter codes for comparison
 It uses mainly three processes:
1. pre-processing:
 discard irrelevant information in input data. Speed, accuracy considered
 Binarisation, normalization, smoothing, denoising done
Online Character Recognition
2. Feature Extraction:
 Out of two or more-dimensional vector field received from pre-processing
algorithms, higher dimensional data is extracted
 Includes pen pressure, velocity or the changes of writing direction
3. Classification:
 Various models are used to map the extracted features to different classes
thus identifying character or words the feature represent
Online conversion
handwritten-to-text Signature Approval Google Translate
Application
Thank you!!!

Handwriting Recognition

  • 1.
  • 2.
    Introduction  Ability ofcomputer to receive and interpret understandable handwritten inputs  Inputs include paper document, handwritten, photograph or from input devices  Input may be “off line” from piece of paper Offline Handwriting Recognition or “on line” from movement of pens Online Handwriting Recognition  “A complete handwriting recognition system also handles formatting, performs correct segmentation into characters and finds the most plausible words.” A segment of handwriting is a piece of the pen-tip trajectory between two defined segmentation points
  • 3.
    Offline Character Recognition Optical Character Recognition(OCR):  Mechanical or electronic conversion of images typed, handwritten or printed text into machine-encoded text  Scanner also scans the texts documents but OCR is different as it scans each letters and characters in the document and saves it in text document whereas scanners scans the documents and saves it as image, not editable
  • 4.
    Implementations of OCR 1.Optophone(1913):  used selenium photo sensors to detect black print and convert it into an audible output which could be interpreted by a blind person  Works on the special behaviour of selenium which conducts electricity differently on light and darkness  As it scans on page it distinguish between dark ink of text and lighter blank spaces  It generates tones corresponding to different letters  E.g. Every vertical line produces a chord then m has three chords and n has two chords  Only a few units were built and reading was initially exceedingly slow  Later models of the Optophone allowed speeds of up to 60 words per minute, though only some subjects are able to achieve this rate
  • 5.
    2. In 1931a machine was developed to convert printed text to telegraph. First one to translate printed text to electrical impulse 3. Optacon(1960s-1970s):  Uses cameras to focus on the letters and replicates the letters on the device and produces 144 vibrating reads which is sensed by index finger
  • 6.
    OCR Techniques 1. Pre-Processing: De-Skew: makes lines of text perfectly horizontal or vertical  Despeckle: removes noises from image and make smooth edges  Binarisation: converts any image to greyscale and separates text from background  Layout analysis: identifies columns, paragraphs, captions  Line and word detection  Script recognition  Normalise aspect ratio and scale
  • 7.
    2. Character Recognition: Usestwo basic character recognition algorithm: a) Matrix Matching/ Pattern Matching:  Compares an image to store glyph on a pixel by pixel basis  Glyph: elemental symbol within agreed set of symbols  Works only if input glyph are separated correctly from rest of image and it matches the stored symbol in similar font in same scale
  • 8.
    b) Feature Extraction: Decomposes glyphs into features like lines, closed loops, line directions and line intersections  These features are compared with an abstract vector-like representation of a character  K-nearest algorithms are used to compare image feature with stored glyph features and nearest match is chose
  • 9.
     Feature detectionin computer vision: computing abstract image information and making local decisions at every image point whether there is an image feature of a given type at that point or not  Software like cuneiform and tesseract uses a two-pass approach to character recognition known as adaptive recognition  Adaptive recognition uses the letter shape recognised with high confidence at first pass to recognise better the remaining letters on the second pass  It is useful for unusual fonts and low-quality scans
  • 10.
    3. Post-processing:  OCRaccuracy can be increased if the output is compared with the letters or words that makes meaningful words or sentence respectively but might not be useful for proper nouns. E.g. I a$e rice can be traced to I ate rice but $am ate rice cannot be traced to Ram ate rice as it could be Pam ate rice  Co-occurrence frequencies can be used by noting certain words seen together. for e.g. “Washington D.C” is seen more often than “Washington DOC”  English grammar can be used to determine whether word is noun or verb
  • 11.
    Accuracy  Study basedon recognition of 19th and early 20th century newspaper concluded that character-by-character OCR accuracy for commercial OCR varied from 81%-99%
  • 12.
    Application  Automatic numberplate recognition  Extracting business card information into a contact list  Make electronic images of printed documents searchable  Assistive technology for blind and visually impaired users  Language translation for e.g. of road signs
  • 13.
     Online inthe sense that inputs are written on the screen instead of printed documents or written paper sheets  Sensor picks-up the pen-tip movements as well as pen-up/pen-down switching  This kind of data is known as digital ink and can be regarded as the digital representation of handwriting  The sensor’s signal is converted into letter codes for comparison  It uses mainly three processes: 1. pre-processing:  discard irrelevant information in input data. Speed, accuracy considered  Binarisation, normalization, smoothing, denoising done Online Character Recognition
  • 14.
    2. Feature Extraction: Out of two or more-dimensional vector field received from pre-processing algorithms, higher dimensional data is extracted  Includes pen pressure, velocity or the changes of writing direction 3. Classification:  Various models are used to map the extracted features to different classes thus identifying character or words the feature represent
  • 15.
    Online conversion handwritten-to-text SignatureApproval Google Translate Application
  • 16.