COBEM99

S22 Mecatrônica

Title:

TECHNIQUE FOR SEPARATION OF OVERLAPPED CHARACTERS

Summary :

ABSTRACT: THE OBJECTIVE OF CHARACTER SEPARATION IS TO SPLIT THE WORDS IMAGE IN MANY PARTS, EACH ONE CONTAINING A COMPLETE BUT ISOLATED CHARACTER. THE MOST DIFFICULT ISSUE IS SEPARATING OVERLAPPED CHARACTERS, BECAUSE THE METHOD USED HAS TO DISCERN PERFECTLY ONE CHARACTER FROM ANOTHER. THE COMPLEXITY IS FOUNDED IN THE DIVERSITY OF TEXT AND IMAGE FEATURES. THE MOST COMMON METHODS OF CHARACTER SEPARATION ARE BASED ON HISTOGRAMS. SOME OF THEM USE VERTICAL PROJECTIONS, OTHERS, VERTICAL HISTOGRAMS. THIS IS THE GRAPHICAL REPRESENTATION OF THE NUMBER OF BLACK PIXELS IN EACH COLUMN OF THE BINARY IMAGE. OTHERS ARE BASED ON HISTOGRAM OF VERTICAL EXTENT, THIS IS THE GRAPHICAL REPRESENTATION OF THE SUN OF THE ALL PIXELS IN THE CHARACTERS IN THE IMAGE. MOREOVER, IT IS BASED ON THE ANALYSIS OF NEIGHBORING PIXELS. THIS PAPER PRESENTS A TECHNIQUE TO SEPARATE OVERLAPPED CHARACTERS BASED ON A HISTOGRAM OF VERTICAL EXTENT, THE GROWTH OF THE REGION BY PIXEL AGGREGATION AND THE ANALYSIS OF NEIGHBORING PIXELS. THE RESULTS CONFIRMED THAT THE DEVELOPED ALGORITHM WAS EFFECTIVE IN THE SEPARATION OF OVERLAPPED CHARACTERS FOR SEVERAL SITUATIONS. KEYWORDS: SEPARATION OF CHARACTERS, PATTERN RECOGNITION, HISTOGRAM

Author :

Ike, Iara Kazuyo
Itiki, Cinthia 0
Okamoto Jr, Jun 0
Silva, Silvana A.Graminho

Paper View :