2D-to-3D conversion of pictures mixed with text and images

초록

Even though a variety of 2D-to-3D conversion methods have been introduced and applied to commercial 3DTVs and 3D displays, the conversion of text regions has not been solved yet. Most of conversion methods generate 3D images with high quality. However, broken or unreadable texts are still visible because the conversion methods are not robust to deal with such non-natural features. Text has the property different from natural images acquired from camera sensors or generated by computer graphics. In other words, due to inherent characteristics of the texts, the best solution is not to convert into 3D, thereby displaying them in 2D. This paper presents a novel methodology of converting non-natural images mixed with texts and images into 3D. For performance evaluation, more than 8,000 images were captured from diverse webpages containing subimages and texts. As well, 3,000 natural images were tested. Experiment performed on those images demonstrates that the detection ratio of non-natural images is 97% and that of subimages reaches 98%. Furthermore, a conversion scheme suitable to non-natural images is proposed. © 2014 ISSN 1881-803X.

키워드