37. READING TEXT FROM PRODUCT PACKAGING

Department: Computer Science & Engineering
Faculty Advisor(s): Serge Belongie

Primary Student
Name: Kai Wang
Email: kaw006@ucsd.edu
Phone: 206-356-6978
Grad Year: 2011

Abstract
In a grocery store environment, there is intuition that along with color and visual design, reading text on product packaging is an important step for humans to perform in identifying items they want to purchase. The field of Optical Character Recognition (OCR) has achieved great success in being able to read scanned documents and books. As we move towards reading text captured in other common situations -- including outdoor signage and text written on product packaging, collectively known as Natural Scene (NS) text -- the task becomes significantly more difficult. Text recognition on natural scene text has become a popular topic and has seen advances in recent years. We propose to investigate this new branch of OCR by using statistical techniques and modern computer vision results, particularly those from the area of object recognition. Our hypothesis is that these new methods are needed to advance the field into a domain of more challenging text. We hope to contribute a new approach to OCR that allows a previously unreadable class of text to be read. In particular, we are interested in performing text recognition to read text from products sold in a grocery store to support our system: a Grocery Shopping Assistant for the Visually Impaired, or GroZi.

« Back to Posters or Search Results