We have recently discovered an easy method for preparing colored Ag films using an aqueous solution of specific sulfide as a coloring agent. The driving force of coloration can be considered to be thin film interference due to the Ag2S layer on the Ag film. On the other hand, electronic papers which have high quality color are widely required. The focus of this present work is to explore the possibility of these color changeable Ag films as rewritable media such as in electronic paper. We found that our Ag films have a potential to be used as rewritable media by controlling preparation conditions.
Edge detection is one of the commonly used operations in image processing. Image processing results depend on the quality of edge detectors. Thus, enhancing edge detectors used on a photo can affect the final result of the application. This paper presents an edge enhancement method based on morphological operations. Edge enhancement is applied on frequently used edge detectors such as Canny, Sobel, Prewitt, Roberts cross, Laplacian of Gaussian and Gabor filter. Pratt's figure of merit is used to evaluate edge detectors results before and after applying the enhancement. Since the enhancement method effects lines in images, nature/ artifact image classification based on line feature is conducted. The results show that the edge detectors, as well as the image classification accuracy got improved. Based on the experiment results, ideal thresholds for each edge detector were determined. Using these thresholds gives high classification accuracy.
In electrophotographic printers and copiers, the fuser fixes resin particles for graphics on sheets by pressing and heating. It is known that curl is generated after the fusing process. One of the main factors of the curl generation is the temperature difference between paper surfaces due to the heating. In recent years, many brands of paper are import or recycling, so that large curl may occur sometimes. We start to develop a prediction method for curl by considering paper properties to describe the characteristics of paper. Permeability is the most influencing parameter of the curl, therefore the permeability was estimated by three methods and the amounts of curl were calculated for ten kinds of paper brands. It is found that the prediction of the curl is approximately corresponding with the experiment when the permeability is estimated from the porosity obtained by ash ratio of paper.
In this article, I give an overview of recent developments of deep learning, focusing on its applications to images. I first introduce new methods for designing and training neural networks that have been proposed since the birth of deep learning and are considered to be standard as of today. I then explain that some recent applications of convolutional neural networks perform even computation similar to global optimization, which is hard to interpret within the concept of classical pattern recognition. Nowadays, owing to these developments, most of visual recognition tasks can be solved by deep neural networks, provided that there is a sufficient amount of training data. Having said that, there is a gap to human vision. To evaluate the gap, I choose and explain a task called VQA (visual question answering), in which, given the image of a scene and a question about it in the form of natural language, we wish to make the computer answer the question. I conclude this article by briefly showing possible future research directions.
With the advances in bio-imaging technology and automatic measurement technology, large-scale biological image data that cannot be sufficiently observed by human eyes has been produced. To acquire biological knowledge from these large-scale image data, researches using classification or clustering of biological images based on machine learning techniques has attracted attention in recent years. In this paper, as use cases of machine learning techniques in biological images, we introduce three research topics : “Automatic classification of biological image data using active learning/semi-supervised learning.” “Single particle analysis for protein three-dimensional structure reconstruction using electron microscopic images.” “Analysis of relationships between genotypes and phenotypes based on large-scale behavioral data analysis of genetic mutants.”
Initially, the effectiveness of CNNs (Convolutional Neural Network) was proved for image categorization tasks for which a CNN accepts an image as an input and outputs a class probability vector as an output in general. Recently the way to use of CNNs becomes diverse, and CNNs which output images have been commonly used for semantic image segmentation, image transformation and image generation. Then, in this article, we explain CNNs for semantic segmentation in case of weakly supervision as well as full supervision, and CNNs for image generation and transformation which are typically decoder-style CNNs and encoder-decoder-style CNNs.
Hand-drawn sketch allows humans to represent visual information in the real world. One of difficulties for computational systems to take hand-drawn sketches is coming from their various type of transformations when we want to compare them with realistic photo images, or consider drawing process. In this article we discuss about deep learning based methods which can take these various types of transformations. First we show how do convolutional neural network based models recognize or generate sketches as static raster images. After that we also discuss other researches which utilize recurrent neural networks in order to process sketches as dynamical processes. In the latter part of the discussion, we briefly introduce out work about integration of visuomotor information in robot's drawing process.
As the question of how to produce good results using the wealth of data brought by IoT has come to be recognized as an important and urgent issue, and analytics in terms of machine learning and AI (artificial intelligence) are gaining more attention. In order to effectively produce good results through analyzing and utilizing IoT data, it is important not only to use advanced analytical methods like machine learning and streaming analytics technology, but also to understand the particular properties of IoT data, and to continuously run the analytics life cycle at a high speed. This paper discusses how to produce remarkable results through analytics utilizing techniques like machine learning, while looking at examples and cases of leading domestic and international companies who have already adopted IoT initiatives.
A lightweight, high stiffness, foldable-deployable Origami structure is very attractive for industry. Origami engineering from Japan in 2003 has great influence to the world and Origami engineering is suddenly coming interdisciplinary. However, the manufacturing cost of complicated origami structure is high, so industrialization of origami is not much advanced. On the other hand, Origami forming, origami printer, folding paper robot shown in this paper also give forming methods suitable for various types of small quantity production according to individual's preference, against the mass production system by mold manufacturing in the 20th century. Describing the present state and issues of origami engineering, it is stated that promotion of the above-mentioned technologies is one of the most important tasks in obtaining the deepening and spreading of origami engineering.