Towards Text Understanding: Word Image Representation, Matching, and Recognition Albert Gordo October 2014
P ART 1 Word Attributes (WATTS) for Word Image Representation and matching
“hotel” “garage” “taxi” “cafe”
?
? “hotel”
“garage” ? “taxi” “cafe”
“hotel” “pizzeria”
“hotel” “pizzeria”
“listen” Lvl 1 […] c d e […] h i j […] l m n […] s t u […] Lvl 2 “lis” “ten” […] h i j […] l m n […] s t u […] […] c d e […] l m n […] s t u […] Lvl 3 “li” “st” “en” […] h i j […] l m n […] […] l m n […] s t u […] […] c d e […] l m n […] 17
Lvl 2 Lvl 3 Lvl 4 Lvl 5 Bigrams [···] [···] [···] er in es […] a b c […] 7 8 9 a b c […] 7 8 9 a b c […] 7 8 9 • •
“hotel” “pizzeria”
Fisher Vector Labeled word images Training Images samples taxi hotel cafe PHOC coffee sushi Training Transcriptions labels
“hotel” “pizzeria”
“hotel” “pizzeria”
ICFHR2014 Competition
P ART 2
• •
Fisher Vector Labeled word images Training Images samples taxi hotel cafe PHOC coffee sushi Training Transcriptions labels
Thank you for your attention
Results: Retrieval
Results: Recognition
Recommend
More recommend