We also evaluate the proposed approach, and the results are quite encouraging. Nevertheless, the proposed approach can dewarp document images having multiple folds. Unlike, the existing methods, the proposed approach can rectify warping in both horizontal and vertical direction. Finally, based on those factors, we dewarp the document image. We estimate those factors from the boundaries of the panels present in a comic document image. Here, we show that warping depends on some factors. First, a simple mathematical model is proposed for warping generation in comic documents. Here, we propose a novel dewarping technique for warped comic document images. So, existing dewarping methods fail to perform with great accuracy in comic document images.
On the other hand, comic documents contain fewer text lines. Most existing techniques work based on the text lines present in the documents. Warping is one of the major problems found on those distorted document images. As a result, different types of distorted images gets generated. People often capture document images using the cameras attached to smart mobile phones. In addition, this paper also collects related CIP datasets, conducts experiments for some typical tasks, and discusses the future work. Thus, this paper reviews the commonalities and differences of 2D CIP methods according to different scenarios and applications, and focuses on recent deep-learning-based algorithms specifically. Especially with the development of deep learning technology, recent CIP methods have achieved better results than direct application of natural image processing algorithms. Therefore, based on the characteristics of cartoons, many specific CIP strategies are proposed. The cartoon images are usually composed of clear lines, smooth color patches and flat backgrounds, which are quite different from natural images. However, there is still a lack of literature to summarize and introduce these 2D cartoon image processing (CIP) works comprehensively. With the rapid development of cartoon industry, various studies on two-dimensional (2D) cartoon have been proposed for different application scenarios, such as quality assessment, style transfer, colorization, detection, compression, generation and editing. Besides, we estimate the errors made by the annotators during the annotation process and describe different evaluation parameters to test the efficacy of the comic document image analysis algorithms. We also elaborate on a couple of applications of the BCBId in the comic research domain. As an application of the dataset, we carry out the sentiment analysis of comic stories-the first-ever attempt on comic book images. A tool is specifically designed for accurate and faster ground-truth generation.
BCBId also includes the metadata encoding of all images in XML format to describe the underlined structure, semantics, and other features of the documents to pursue research on understanding stories and dialogues. BCBId has the ground truth for extracting various visual components of the comic book images, i.e., panels, characters, speech balloons, and text lines. Bangla is the 6th most popular spoken language in the world-used by 265 million people (), and has a century-old heritage of comic strips (in newspapers) and books.
BCBId consists of 3327 images taken from 64 Bangla comic stories written by 8 writers. This paper describes the creation of the first-ever comic dataset among Indian Languages, namely Bangla Comic Book Image dataset (BCBId) (), which is also made public for the benefit of the researchers. However, comic document image processing research suffers due to its inherent complexities and the limited availability of benchmark public datasets. Comic document image analysis is now an active field of research in both academia and industry.
0 kommentar(er)
