Computer Vision Ronald Frazier CIS 479 April 20, 1999 Introduction • Important area of study – Ease of use for new users – Managing large quantities of images Combines a Variety of Disciplines • Standard Programming Techniques • Artificial Intelligence Techniques – Neural Networks – Fuzzy Logic • Image Processing Techniques • Human Biology and Physiology Optical Character Recognition (OCR) • Converting graphical representation of text to character based representation • One of the most developed areas of computer vision • Lots of ideas for further research OCR - Character Recognition • Examining graphical representation of character to determine character represented • Identification Process – Apply image processing – Identify character (usually using neural networks) OCR - Sample Character Recognition Algorithm • Chaincode Algorithm – Apply image processing to get outline – Break character into 4x4 grid – Calculate average slope and curvature for each cell OCR - Sample Character Recognition Algorithm • Chaincode Algorithm – Data processed by Neural Network – Character recognized by Neural Network Types of OCR • Preprinted Text • Live Handwriting OCR - Preprinted Text • Text that has already been written or typed before recognition begins – ex: Encyclopedia, Contract, Report • Can be used on printed or handwritten documents OCR - Preprinted Text • Recognition Process – Identify possible individual characters • Standard character recognition techniques – determine possible words – Compare to dictionary and determine existing words – Select most likely existing word Applications - U.S. Postal Service • Handle over 100 billion parcels yearly. • Need automated way to identify destination of each parcel and print a barcode for faster processing. • For more details, see my web site OCR - Live Handwriting • Process text as it is written – ex: Personal Digital Assistants • Can be used on handwritten documents • Track time, direction, and angles of lines as they are written and use it to identify character Content Based Image Recognition (CBIR) • Identifying contents of images • Applications – Automated organization and classification of images – Image database searching – Still images or video CBIR - Recognition Methods • Content that can be recognized: – Specific colors and approximate proportions • ex: A lot of Red, a little bit of Green – Objects base on shapes, colors, textures, edges, size, etc. – Face detection Applications - News Video Recognition and Retrieval • Store video in database along with description of content for searching • Automated determination of video content Applications - News Video Recognition and Retrieval • Processing Technique – locate and extract on screen captions and closed captions – Apply OCR to convert captions to text Applications - News Video Recognition and Retrieval • Processing Difficulties – Interference from background image • Handled by Frame Filtering and Frame ANDing – Low resolution of captions • Handled by magnifying characters and interpolating pixel values Other Possible Uses of Computer Vision • Robots • Self-Controlled Vehicles • Security - Fingerprint and Retina Matching Additional Information • For more information on additional topics not covered in this presentation, along with links to other computer vision pages, see my web site at: http://people.mw.mediaone.net/ldkronos/AI