Recognizing Thousands of Legal Entities through Instance-based Visual Classification

Abstract

International audienceThis paper considers the problem of recognizing legal en-tities in visual contents in a similar way to named-entity recognizers for text documents. Whereas previous works were restricted to the recognition of a few tens of logotypes, we generalize the problem to the recognition of thousands of legal persons, each being modeled by a rich corporate identity automatically built from web images. We intro-duce a new geometrically-consistent instance-based classifi-cation method that is shown to outperform state-of-the-art techniques on several challenging datasets while being much more scalable. Further experiments performed on an au-tomatic web crawl of 5,824 legal entities demonstrates the scalability of the approach

DigInPix

“…The only achievement dealing with more realistic numbers of entities and richer visual identities is the recent work of Leveau et al [7]. The authors actually reported in that paper an experiment with more than 5,000 entities represented by a (noisy) training set of 370K images automatically crawled through a web search engine.…”