Extração autumática de texto em vídeo

Automatic Text Extraction in Digital Video based on Motion Analysis Duarte Palma, João Ascenso, Fernando Pereira

Instituto Superior Técnico – Instituto de Telecomunicações, 1049-001 Lisboa, Portugal e-mail: {Duarte.Palma, Joao.Ascenso, Fernando.Pereira}@lx.it.pt

Abstract. It is well known that the text that appears in a video scene or is graphically added to it is an important source of semantic information for indexing and retrieval, notably in the context of video databases. This paper proposes an improved algorithm for the automatic extraction of text in digital video; its major strengths are its robustness in terms of text skew and its improved performance in dealing with scene text. The system is based on a segmentation approach, using geometrical and spatial analyses for text detection. After, temporal redundancy is exploited to improve the detection performance by means of motion analysis. The output of the text detection step is then directly passed to a standard OCR software package in order to obtain the detected text as ASCII characters.

Introduction

The technological advances seen in recent years in the area of audiovisual representation technology have led to a boom in the usage of audiovisual information, namely accessed through the Internet, by a growing number of users. The increasing amount of audiovisual information being deployed has led many relevant content players such as audiovisual content producers and television operators to show interest in creating digital libraries which should allow the efficient storage and indexing of audiovisual information for future management and retrieval. Nowadays, the task of annotation is typically performed manually, by a human operator; this process is very expensive, time consuming and many times suffers from the subjectivity associated to the human operator. To address the need and overcome the problem, it is necessary to develop systems capable of automatically processing