Research On Logo Detection And Recognition In Document Images

Logos, widely applied and also referred to as trademarks, are visually-salient symbols which appear on manufactured articles and serve as remarkable identifications of related organizations. In document analysis domain, the primary concern of the logo is to use it as an effective assistant way of identifying the source of a document. This paper discusses the topic of utilizing logo for document retrieval, which basically involves two aspects: (1) detecting the probable logo from a document image with required precision and accuracy; (2) indexing the detected logo candidate region into a database to classify it or conclude that the detected logo is not of interest. The first matter is referred to as logo detection and the second is called logo recognition. In our work, we initially reorganize the structure of the logo detection and recognition procedures and integrate them into a unified framework. Experiments show that the proposed method outperforms existing methods in document processing domain.