Tomo Miyazaki

Ultra-low resolution character recognition system with pruning mutual subspace method

Improvement of character recognition technology brings us various character recognition applications for mobile camera. However, many low-resolution and poor-quality character images exist due to the performance of the camera or the influence of environment, and existing methods are not good at recognizing those low-resolution characters. Therefore, we develop a character recognition system for ultra-low resolution character images less than 20*20 pixels. The proposed system consists of three phases: increased training data with a generative learning method, creating a deblurred high-resolution image with Wiener filter and image alignment, and recognition by pruning Mutual Subspace Method.

Estimation of gazing points in environment using eye tracker and omnidirectional camera

In this work, we propose a method for estimating the users gazing point in the environment using images taken by an eye tracker and an omnidirectional camera. The proposed method estimates the eye positon in environment by mapping the gazing point obtained by the eye tracker in the omnidirectional camera image. However, matching the omnidirectional image and the eye tracker image is difficult because the omnidirectional image is distorted by equirectangular projection. Therefore, we propose a method for estimating eye location in the omnidirectional image by matching the eye tracker image to the omnidirectional image with considering the distortion. Specifically, this method repeats image matching and image conversion using the matching results.

Activity Recognition Using Gazed Text and Viewpoint Information for User Support Systems

The development of information technology has added many conveniences to our lives. On the other hand, however, we have to deal with various kinds of information, which can be a difficult task for elderly people or those who are not familiar with information devices. A technology to recognize each person’s activity and providing appropriate support based on that activity could be useful for such people. In this paper, we propose a novel fine-grained activity recognition method for user support systems that focuses on identifying the text at which a user is gazing, based on the idea that the content of the text is related to the activity of the user. It is necessary to keep in mind that the meaning of the text depends on its location. To tackle this problem, we propose the simultaneous use of a wearable device and fixed camera. To obtain the global location of the text, we perform image matching using the local features of the images obtained by these two devices. Then, we generate a feature vector based on this information and the content of the text. To show the effectiveness of the proposed approach, we performed activity recognition experiments with six subjects in a laboratory environment.

Analysis of floor map image in information board for indoor navigation

Various indoor navigation methods have been developed recently, but digitalized data of indoor map is not always available. Therefore, an indoor navigation framework using an image of information board has been proposed. In this method, the process to extract map regions from the image of an information board is necessary to be done by hands beforehand, and the process to estimate passageway regions is important because its information is used in map matching. However, the method of passageway discrimination is very heuristic, which is intended for a specific type of floor maps. Therefore, in this paper, we propose a semi-automatic method to extract map regions from the image of information board with simple users operation. We use GrabCut method and Snakes method for the extraction method. In GrabCut method, we detect closed regions to prevent the degradation of accuracy when conducting GrabCut to the downsizing image. The proposed method can extract a map region with few deficits in short calculation time. In addition, we propose a machine learning based method to classify passageway regions and other regions from a segment image. We confirmed that the proposed methods are effective and promising by experiments.

Text Retrieval for Japanese Historical Documents by Image Generation

Digitization of historical documents is growing rapidly. Text retrieval is a vital technology to facilitate the use of historical document images because of the large amount of data. In this paper, we propose a method for retrieving keywords in Japanese historical documents with text query. The proposed method automatically generates an image of the query text and retrieves regions in documents similar to the generated image by feature matching. We exploit a technique of deep learning to generate an image close to texts in Japanese historical document images. Furthermore, we use convolutional neural network to extract features robust to appearance variation of texts in documents, such as shade and shape of texts. We conducted the text retrieval experiments on the public dataset of Japanese historical documents in the Edo era. The experimental results show the effectiveness of the proposed method.

Object-Based Video Coding by Visual Saliency and Temporal Correlation

When a disaster occurs, video communication is an effective way to disseminate large quantities of important information. However, video coding standards such as High Efficiency Video Coding (HEVC) compress entire videos, whatever the contents are; at low bit rates, the quality of significant objects deteriorates. In this paper, an object-based video coding method is proposed to address this problem. The proposed method extracts objects on the basis of visual saliency and temporal correlation between frames. Subsequently, we execute pre-processing which degrades the background quality before encoding the video with HEVC. This method can reduce the bit rate while preserving target object quality. Experimental comparison with HEVC demonstrates the superior performance of the proposed method.

Graph model boosting for structural data recognition

This paper presents a novel method for structural data recognition using a large number of graph models. Broadly, existing methods for structral data recognition have two crucial problems: 1) only a single model is used to capture structural variation, 2) naive classification rules are used, such as nearest neighbor method. In this paper, we propose to strengthen both capturing structural variation and the classification ability. The proposed method constructs a large number of graph models and trains decision tree classifiers with the models. There are two contributions of this paper. The first contribution is a novel graph model which can be constructed by straightforward calculation. This calculation enables us to construct many models in feasible time. The second contribution is a novel approach to capture structural variation. We construct a large number of our models in a boosting framework so that we can capture structural variation comprehensively. Consequently, we are able to perform structural data recognition with the powerful classification ability and comprehensive structural variation. In experiments, we show that the proposed method achieves significant results and outperforms the existing methods.

Efficient Coding for Video Including Text Using Image Generation

Text in video compressed by lossy compression at a low bitrate will easily be deteriorated, resulting in blurred text and a lower readability. In this paper, we propose a novel image coding method to preserve the readability of text in the video at a very low bitrate. During the encoding process, we estimate the parameters for each character of the text. Then, an image without text is generated and compressed. During the decoding process, we reconstruct video sequences with text from images without text and character images generated by the estimated parameters. The experimental results show the effectiveness of the proposed method in terms of the readability at a very low bitrate.

Adaptive Post Filter for Reducing Block Artifacts in High Efficiency Video Coding

This paper describes an adaptive deblocking postfilter based on neural networks for use in H.265 High Efficiency Video Coding (HEVC). Blocking noise is a common problem in video coding caused by the division of the frame into blocks. The filter is adaptive because it uses different filter parameters depending on block characteristics. We use a modified HEVC decoder to export the block information to the filter. Blocks are put in different categories according to transform and prediction parameters. We train the filter in each case to optimize the internal parameters. We find that a 2-layer convolutional neural network is able to outperform the in-loop deblocking filter for a moderate processing cost. We propose to apply the filter after the HEVC in-loop deblocking filter. We demonstrate that our filter helps further reduce visual artifacts in video.

Fast method for extracting representative graph from decorative character images

In this paper, we propose a fast method for extracting a representative graph from decorative character images. The representative graph is an effective method to describe structure information of a graph class. However, the computation time of existing methods for extracting a representative graph is extremely expensive. They search all combinations of nodes of a class. The proposed method calculates the common nodes in a class which have same features before the representative graph is extracted. The computation time can be reduced using the common nodes. To show the validly of the proposed method, experiments are carried out using decorative character images.