In recent years, in offices, there is an increasing need to manage and operate not paper documents but electronic documents in order to use and cooperate with the cloud and RPA (Robotic Process Automation), and a large number of paper documents can be computerized easily and optimally, the importance of capable functions is increasing. On the other hand, in digitization of paper documents, it is difficult to optimize the image quality and the size. If the image quality is prioritized and scanned in gray scale, the file size becomes large. In addition, if you give priority to size, and do not use grayscale but scan with black and white, the distinction between text and background will be unclear, so there is also a problem that the character can not be visually recognized. Therefore, we have developed a new technology that automatically determines an appropriate halftoning for each document from either black and white or gray scale. In this paper, we report on the outline of this technology.

Metrics which can quantify perceptual differences between appearances of two objects is highly useful for developing photographic reproduction systems because such metrics can be used as a performance index of them. Meanwhile, appearance of material surfaces is well studied when it varies in a single attribute such as colors, glosses or textures, while the appearance of material surfaces in real world consists of mixed perceptions yielded by those attributes.

In this study, we aimed at quantifying the perceptual differences in unified appearance of objects using CG-faces of different colors and glosses. Our prediction used up-to-third order image statistics and showed high correlations with the subjective responses. Furthermore, the subjective responses to the face stimuli show strong correlation with the subjective responses to abstracted stimuli which differ in pixel alignments only. This strongly suggests that the metric has some generality and it is not limited to the facial appearance prediction.

Manga acts the central role of Japanese entertainment industry of visual arts just now. And it influences the young peoples all over the world. Our country have a long history of making narrative works that combine pictures with letters. It began as the Emakimono (picture hand-scroll) made in the court of Heian period. In later Edo period, according to the development of the wood-cut printing, the Ezoushi (popular illustrated book) flourished as popular culture along with the Ukiyo-e printing. In the same period, Katsushika Hokusai used the term “Manga” on the title of his best-selling series of illustrated books “Hokusai Manga. ” This essay gives an overview of some attempts to make narrative works that consist of pictures with letters, not only the change of the meaning of a “Manga” word, from these old periods to early Showa period through the civilization era of Meiji period. Today's many elements of Manga came from not only domestic factors but from western influences.

Manga face identification is important for content-based manga processing. However, it is a difficult task because manga character drawing styles differ greatly among artists. To accurately cluster faces within an individual manga, we propose a method to adapt manga face representations to an individual manga. We use deep features trained for generic manga face recognition, and adapt them by deep metric learning (DML) for the target manga volume. DML uses pseudo positive and negative pairs defined by considering page and frame information. We performed experiments using a dataset comprising 104 manga volumes and found that our feature adaptation significantly improved the performance of manga face clustering and we obtained 0.08, 0.16 improvements in NMI (Normalized Mutual Information) and accuracy, respectively.

Extracting information from comic images is more difficult than that from natural images such as photographs. This is because they are expressed in black and white line drawings and the contents are often drawn using special techniques such as exaggeration and simplification. Due to these characteristics, sufficient extraction accuracy cannot be obtained using general functions of conventional image processing such as SIFT (Scale-Invariant Feature Transfer) and HOG (Histograms of Oriented Grandients).

In recent years, however, techniques of applying deep learning to comics have been proposed, contributing to a significant improvement of accuracy.

In addition, this technology can be used interpreting and estimating comic stories. This paper introduces research trends in deep learning applied to comic image processing.

After the '90s, the research focusing on Manga has been progressing in informatics. It is expected that the development of AI technology enables advanced image processing to deal with semantics and contents in manga. Metadata is useful to handle contents in the image of manga. This paper introduces some research of metadata for the use of contents and their structure included manga. This refers to our system to provide images of manga with its metadata as annotations. To develop this system, we utilize International Image Interoperability Framework (IIIF) which is an international standard to share images and its annotations.

In the field of image engineering, this paper introduces that four-scene comics is the one of the most attractive media, which contains complex problems. Especially, four-scene comics is multi-modal contents which is composed of images and languages, the length and the shape is limited so that there has stereotyped patterns. Despite the numbers of scenes is shorter than other types of comics, the comic artists can draw various kinds of stories. Thus, the artists should select appropriate complex objects to draw emotions and background knowledge in daily scenes, cross-domain technologies are required to comprehend stories. In other words, the researchers can find novel technologies by focused on these contents. This paper describes the my research project of “Developing a Dataset and Framework to Analyse Stories in Four-scene Comics Using Deep Neural Networks” adopted as one of projects of JST ACT-I in 2017.

This paper proposes a system named “Zugaan” that supports to create cartoon plots through overviewing the study on comics. Varied contents exist in comics and those can be assumed as the part of the culture. Comics can be recently not only assumed as just entertainment contents but also used as a communication tool with widespread social media. In the proposed system, we aim to support two aspects of activity related to comics : creating and research. The proposed system enables users who are not good at drawing to design the points that professionals concern by the intuitive interface. Even for professionals, what types of comics can effectively show their intentions should be not sure. Several kinds of research have analyzed a lot of comics to find such kinds of specific features entertaining the readers. However, those results greatly depend on the effectiveness of image processing technology. The comic plots created by the proposed system can be with metadata and should be useful for analyzing comics from the research aspects.