Sei sulla pagina 1di 13

Dr.

Mallikarjun Hangarge
Document Image: Why?
• Paperless Solution
– Efficient transfer
– Organization
– Convenience
• Access to a variety of content
– Universal reader – email, attachments, spread
sheets
– Don’t need original applications
How Do We Acquire Document Image?
• Scanner
• Camera
• Smart Phones
Where we find ?

Everywhere
What we can do with them?
• Can we Access it?
– Search
– Browse
– “Read”
• Index and Retrieve them?
In their basic form not really!
• We can
– View
– Print
– Not much else
Why?
1. Image ID
Query
2. Structure
Documents
3. Decomposition
4. Handwriting
Layout
Similarity
Ranked 5. Stamps/ Logos
Results
6. Zone Classificatio
Images
w/Text
Genre Class
Classification Results

Page Document Handprint Line


Enhancement
Classification Images Detection

Hand
Signature
Noise Page Detection
Decomposition

Images Zone
Machine Segmentation
w/o Text Labeling

Stamp and Logo


Graphics Detection

< .5 .25-3 1-3 1-3


Target Processing Speed in Seconds
OCR
• Limited OCR
– Extensive History ( someone else work)
– Engineering Oriented ( Hindi OCR, Bangla OCR)
– Needs significant Normalization
• Degraded Imagery
• Our Priority
– Unknown heterogeneous content
– Graphics( Logo/ Seal) and Handwritten content
• Focus on Development of system for Classification and
Retrieval of Heterogeneous Documents
Traditional Approach
• Conversion/OCR + Text Retrieval
– Advantages :
• Gets support from prior Information retrieval research
• Text Retrieval optimized using against relevance
– Disadvantages : Text retrieval is only as good as OCR
• OCR accuracy varies widely
• Poor quality of Images or Unique fonts
• Character, word or line segmentation is hard
• Cant Process Graphics Objects
• Page Structure Difficult to Preserve
Why (document) image retrieval ?
• Wide use of scanners created Huge document
image collection:
– Historical documents : books, letters, Manuscripts
– Paperless Office : Memos , Letters, Notes, forms etc..
– For this all Images are common format
– Increase in amount of content
• Research challenge :
– How to make this content available in searchable
format for users ?
Current Trends in Document Image
Retrieval
• Search directly against pixel content :
– Page structure : Layout Analysis/ classification
– Graphical Objects : Logos, Signature, seals
– Named Entity , Date field
– Word Spotting
• Drawbacks
– Don’t allow free form queries
– Algorithms tasted on limited data sets and don’t scale
– Technique is not compared to text retrieval
– No Techniques tested for user relevance
– Most techniques use training and recognition
Large Scale Image access
• Retrieval :
– Features
– Indexing
– Querying and Retrieval
• Our Contributions
– Testing potential of Global and Local Attributes for
Word Retrieval from Kannada Documents
– Logo Detection in Document Images
Word Spotting
Logo Detection and Retrieval

K C Santosh et al. 2015

Potrebbero piacerti anche