ABSTRACT

Contents 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96 4.2 Properties of Content and the TV Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . 97 4.3 Detection of Overlaid Text by Classifier Fusion . . . . . . . . . . . . . . . . . . . . . . . . . . 98

4.3.1 Edge-Based Preprocessing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 4.3.2 CC-Based Character-Level Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100 4.3.3 Texture-Based Text Detection by SVM.. . . . . . . . . . . . . . . . . . . . . . . . . . 101 4.3.4 Text Verification by a Decision Tree. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 4.3.5 Text Detection Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

4.4 Detection of TV Channel Logos by Using Cinematic Cues . . . . . . . . . . . . . . 107 4.4.1 Computation of the Scene Model by Color. . . . . . . . . . . . . . . . . . . . . . . 108 4.4.2 Identification of the Outliers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 4.4.3 Extraction and Verification of Logos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

4.4.4 Logo Detection Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110 4.5 Applications for TV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112

4.5.1 Enhancement of TV Graphics for Improved Visual Quality . . . . . 112 4.5.2 Simultaneous TV Burn-In Protection and Text Contrast

Enhancement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114 4.5.3 Modification of Text for Improved Readability. . . . . . . . . . . . . . . . . . . 116

4.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119

Overlaid graphics, such as text and channel logos, are extensively used in television (TV) programs to provide additional information to the audiovisual content. In the literature, their automatic extraction has mainly been considered in non-TV applications, such as for automatic video indexing and retrieval. In this chapter, we propose algorithms for automated detection of overlaid graphics with limited computational andmemory resources of a TV set. Specifically, we deal with text and logo detection, and introduce novel applications that use their results for visual quality improvement, contrast enhancement during burn-in protection, and customizing text appearance for increased readability.