LOGO DETECTION IN DOCUMENT IMAGES

Abstract

This paper focuses on the problem of classifying segments of bitonal document images according to whether or not they are likely to contain a logo. A simple set of features is extracted from each segment, based primarily on statistics about the connected components of black pixels within the segment. The features are then used to construct a set of rules to determine the classification. 1 Introduction This paper describes one part of a larger project whose goal is to analyze electronic documents and to determine their source based on an identifying logo. We divide this problem into three basic steps: Segmentation The document is partitioned into smaller images, which contain logically related portions of the document (e.g., small text, picture, large text, logo, etc.) Logo Detection Features are extracted from the segments which are then used to classify the segments based on whether or not they are likely to contain a logo. Logo Matching Segments which have been identified as like...