The construction of training datasets for machine learning to estimate the image impression of generally requires annotating a large number of images. Here, it is problematic in that the training results depend on the individual impression responses of the operators. In this study, we propose a semi-automatic system for annotating impressions of images based on the results of machine learning of the impression responses of many operators, and a method for visualizing the annotation process in order to reduce the individual differences in the impressions of images and the burden on the operator. First, the system evaluates the impressions using the SD method, and then generates a fuzzy decision tree using the impression values of each image. The fuzzy decision tree automatically classifies images, and then visualizes the results and the process of classification to assist the user in reclassifying images. By linking the display of the decision tree with a similarily-based image browser, we improve the readability of the decision tree and support understanding of annotation trends. In this paper, we present an example of visualization using this method on data from 43 workers' impression evaluations of 1,500 images of women's clothing, and verify the usefulness of this research through user experiments.

View full abstract

We design and implement tools to support the creation of storytelling mazes. While the conventional process of creating storytelling mazes has been done on static media such as paper, this study develops computer-based tools and discusses their feasibility and usefulness. The process of creating storytelling mazes is complex, the creator have to make stories in parallel with the construction of the maze. Based on this complexity, we present a direction for creation support, and design interactive tools. Specifically, we propose a user interface that allows users to create works by moving back and forth between three blocks: making a story, placing it on the maze, and changing the maze parameters. We also propose a method to solve the maze placement of the story as an optimization problem by using a stochastic algorithm. We show some results actually created using the proposed tools and discuss usefulness.

View full abstract

The stone tools excavated from the ruins are washed, given identification number, classified for assembly, and then assembled. After that, each stone tools are investigated to know its manufacturing method and usage. However, since the identification information may be lost in the process of investigation, a system that automatically presents the identification number from the actual stone tool is required. This paper propose a method for identifying real stone tools measured by an RGB-D camera in order to automate the identification of stone tools. In this method, the point cloud of the stone tool measured by an RGB-D camera, and the point cloud that has been scanned in advance are compared by an algorithm that combines 2D and 3D methods. As the result stone tool becomes identified. Specifically, by combining the matching method using 2D contours and the ICP algorithm, we have achieved faster processing and higher discrimination accuracy. As a result of creating a database of 82 stone tool point cloud data constructed from 24 stone tools and applying this method to some stone tools, good results were obtained.

View full abstract

In this study, we propose a tracking method for dynamic projection mapping (DPM) on living things. We also aim to report the creation of a video artwork with projection mapping silkworms, hedgehogs, and mushrooms as projection objects in this paper. In most of the previous researches and works on DPM for non-human deformable objects, the 3D shape of the projected object was input in advance or markers for tracking were attached to the projected object. In addition, there are no studies of DPM for animals and plants. In this method, we apply a common tracking method to silkworms and hedgehogs by tracking their end points. In addition, we propose an effect animation based on the shape of each object with the concept of bioluminescence.

View full abstract

An aesthetical image composition and mysterious stories are the main components of a hidden object game (HOG) to make it enjoyable. In an HOG, players find objects hidden in an image and inquired by a list. As far as we know, there have been insufficient studies to explore game elements such as the efficiency of game effects on an HOG to accelerate human activities. In this paper, we propose "wiping operation" to accelerate human activities for an HOG. In an HOG that we developed, first, an original image is transformed into a triangulated image. Then, during game play, the quickness of the wiping action by a player makes the level of detail of the triangulation finer and rougher. We did an experiment to investigate the performance of the wiping operation. Each player in the experiment wore a brain activity sensor to obtain a brain activity value during game play and answered a questionnaire after game play. We show the experimental results by the questionnaire and brain activity value to verify the effectiveness of the wiping operation.

View full abstract