We propose a method for detecting L-shaped layout and superimposed text from broadcast video. The method detects L-shaped layout by analyzing edge information and variance of pixel values, and then converts text image into characters using OCR. Experimental results showed the effectiveness of the proposed method.

View full abstract

In this paper, we propose the tracking system of human beings with time-of-flight range camera. Each pixels of image from camera have information of distance between camera and object. In this method, each objects are detected and tracked independently of temporary occlusion.

View full abstract

This paper studies a method of detecting the window areas on buildings from web bird's view images. Unlike conventional methods, our method does not depend on the edges of buildings, but utilizes regular patterns of textures of windows to cope with the low resolution of images and window reflection. Experimental results show that our method can precisely estimate window area ratio on the walls for 73% of all buildings.

View full abstract

Robustness of ear recognition system of [1,2] is improved.

View full abstract

For the situation having difficulty in localization with the GPS, this paper proposes of localization method using cell-phone. In addition, this paper studies the influence of noises for this method caused by the difference between original database motion pictures and unexpected blurred still pictures of cell phone.

View full abstract

We present an image processing system for counting people, which use simple linear regression and multiple linear regression from low level image features. The low level features include statistics of R, G, B, number of edges, etc. The result show good estimation for real scenes.

View full abstract

We developed a method to segment food images automatically. In this method, we detected local extrema of luminance and applied GrabCut on their convex hull. The recognition rates were 90.38% for food regions and 68.12% for non-food regions.

View full abstract

This paper experimentally studies a variance-stabilization approach to denoising low-light observation color images, which are often referred to as Poisson random color images, and in denoising performance compares the variance-stabilization approach with another denoising approach utilizing noise's signal-dependence directly for denoising.

View full abstract

We present a technique for suppressing noises in image sharpening method using the bilateral filter with image decomposition into a base and detail components. The detail component is smoothed with a cross-bilateral filter then magnified and added to the base component to get a sharpened image. We demonstrate that this technique outperforms a conventional method with preliminary smoothing of input images with respect to denoising and sharpening capability.

View full abstract

We propose a method for enhancing the contrast of color images by transforming the original colors into the colors on the boundary of RGB color space. Experimental results show that the proposed method generates colorful and contrast-enhanced images.

View full abstract

We developed a video distribution system that adjusts the video encoding rate according to client number on the IP network. This system can keep to distribute live videos without interruption in case of sudden increasing access. We implemented the system, and affirmed that distribution rate is falls inside maximum bandwidth of the server.

View full abstract

We measured the electric field strength of FM broadcasting signals on the train and predicted the coverage area for mobile reception of mobile multimedia broadcasting signal in the VHF-Low band on the train. As a result of the measurement, the penetration loss of train was estimated about 10 dB and it was shown that required electric field strength out of the train was 59 dBμV/m.

View full abstract

We studied on the relation between the time-interleave length and correct reception ratio in the mobile reception of mobile multimedia broadcasting signal in the VHF-Low band. As a result of the laboratory experiment, the required received input power of the receiver with the time-interleave length I=8 was 1.5 dB lower than that of the receiver with the time-interleave length I=4 at the maximum Doppler frequency (Fd_max) of 2 Hz.

View full abstract

In the storable service, the data carousel transmission method is proposed. This method sends and receives storable contens by the unit of DDB. So we clarified the relationship between DDB size and transmitting efficiency of storable contents.

View full abstract

This paper describes the method of spectrum sensing based on feature detection of ISDB-T signal. To detect a signal that is smaller than the electric power of the thermal noise, we propose the technique for synchronously adding the SP signal regularly sent. We developed experimental device and evaluate the performance of this method.

View full abstract

This report addresses the problem of channel estimation of ISDB -T signals, especially in the receiving environment with many reflections. We propose the adaptive control method of leakage in time domain channel estimation and its appication to the frequency domain equalizer. Experimental results show the effectiveness of the proposed method.

View full abstract

We have proposed a sequential color system for use as a video format. To realize this system, a conversion to whole RGB picture with the interpolation of missing color plane is important. Motion estimation methods for the motion compensation using the contour pictures of different color plane are proposed.

View full abstract

We have proposed the sequential color system as a video format with which R,G,B are changed by a frame. To realize this system, it is important to predict the lacking color without motion compensation. In this time, we consider prediction method in the case of one color lacking. Pixel Value is predicted by existing colors using the color probability of the reference picture at each cluster.

View full abstract

In the many kinds of video system, it is desired that the picture quality is controlled to provide the better picture. For this purpose, an objective picture quality estimation method without the reference picture is required. To respond this, a spatial blur estimation using DCT is proposed.

View full abstract

Sequential color has only a plane in a frame and each planes are switched for each frame. To realize this system coding efficiency is important. Therefore, the coding efficiency of luminance/chrominance system and RGB system are compare at Intra coding.

View full abstract

We successfully developed an optical high-speed gaze controller, called the "Saccade Mirror", to achieve a high-speed pan/tilt camera with a high-speed image processor. In this paper, the new prototype system which can be used with a Full HD image sensor is proposed.

View full abstract

Orthogonal transformation coding method has been used as a popular technology in video compression field. And almost of them have sanctioned the efficiency of the Discrete Cosine Transform (DCT) in particular. On the other hand, other transforms like Discrete Sine Transform (DST) and Discrete Wavelet Transform (DWT) have been paid more attention. This paper addressed the efficiency diversification of each kind of transform when the quantization step or input image is changed.

View full abstract

A function has been developed to detect the 3D position of a finger-tip in real time using only one camera. This function can be added in the finger-tips 2D mouse system to realize a finger-tips 3D mouse.

View full abstract

Increasing demands to utilize large-scale video properties such as broadcast archives, video search technologies to find intended scenes effectively are needed on TV productions. Currently, keyword search methods are mainstream, based on text information assigned manually to each scene in videos. However, needs of video search for program creators are vary widely, so there are search demands which are difficult to express using only keywords. This time, we report about a prototype of program video search system with multiple faculties based on our media analysis technologies (image analysis, natural language analysis). We aim to put this prototype into use as a broadcast archives search system on TV production fields.

View full abstract

We propose a system for recognizing horn and siren sounds. We calculate the degree of harmonic structure from the amplitude spectrum and search its peaks. The peak frequency corresponds to the fundamental frequency of warning sounds. The experimental results show the method can detect warning sounds in SNR of 0dB.

View full abstract

This paper presents a detection method of important events based on relationship between player action and sound on stadium in soccer videos. Generally, since player action and sound on stadium have high correlation in the important events, we realize the detection by using their relationship based on canonical correlation analysis.

View full abstract

This paper proposes a color estimation method of team uniforms in soccer videos based on color co-occurrence. The proposed method calculates color correlograms and selects sets of color components, whose co-occurrences are higher than the other sets. This enables the color component estimation of the team uniforms, which contain multiple colors.

View full abstract

Pathological diagnosis support system extracts positions and contours of hepatocellular nuclei semi-automatically, and gives pathologists useful information such as nuclear density and roundness factors of nuclei. Accuracy for nuclear position and contours extraction has been improved to reduce the time for the pathologists to collect wrong results.

View full abstract

The authors is developing M[you]sic: an interactive music system for city planning. In this paper we report implementation of image input for M[you]sic. It will contribute to enhancing application of M[you]sic and gathering data for city planning. We also report the experiment of the system which was carried out in Maebashi.

View full abstract

We realize HDR display by multiple projection using four projectors. However, response functions of the HDR display are not always suitable for all images to display. Therefore, this study examines response functions of the HDR display by considering the contents of the images.

View full abstract

Our research objective is to judge document similarities using Orthogonal Transformation. We focus on Fourier system in this paper from the results of previous experimentation. In previous experimentations, what single-byte characters and double-byte characters are intermixed was a problem. Therefore we propose the characters-code-conversion against this problem. Precision of the judge rose by the characters-code-conversion, which standardizes characters code to double-byte characters. Furthermore, we decided threshold of each length of sentences.

View full abstract

In this paper, we estimated the calorie of food in the pictures using features extract from food images. We use multiple image features such as color histograms, color correlogram and SURF. We firstly estimate using only one feature, then we use multiple linear regression analysis and weighted the estimated values. We used 6996 ordinary food images from web service "FoodLog". The distribution of the estimation shows that 76.2% of the estimations are correct within ±40% error and 48.0% correct within ±20% error.

View full abstract

A face authentication method using SURF with ASM Guide is proposed. The recognition performance and robustness to face direction of our proposed method are discussed. It is observed that the face turn reduces the recognition rate of our method.

View full abstract

We propose a novel face recognition system against the various expressions. Our proposal method needs only one neutral face image and produced various expression face images which are registered in the training database from one of them. The results show that four expressions by using parameters obtained can be made.

View full abstract

The Difference of Gaussian-based illumination compensation is presented to provide robust eigenface-based face recognition against large face variability due to lighting variations. This work adopts the Difference of Gaussian from SIFT for filtering the illumination component of face image. The results show the proposed method could solve the illumination problem.

View full abstract

We have studied difference between global PCA and short term PCA of motion pictures for online rhythm evaluation of dance. We have found that short term PCA with at least twice the length of cycle of motion enable us to extract rhythm features of dance as accurate as global PCA in most cases.

View full abstract

This study examined the differences of spatiotemporal kinetic information between participants' original motion and imitating motions of two models. The side of models' hands, ipsilateral/contralateral hands, was utilized as an experimental parameter. The result indicated that imitating contralateral hand is more likely to adjust their body spatially rather than temporally.

View full abstract

We will briefly describe several methods for measuring the crowd and the number of people for the sake of controlling contents on digital signage systems based on the people distribution in a target area.

View full abstract

A control system is proposed for home appliances using only slight hand-waving gesture. The system detects the gesture with four cameras and calculates its 3D-position. Some operations of TV (turn on/off, channel change, volume up/down, etc.) can be related the hand-waving positions. Experimental results show that the system is reliable.

View full abstract

This paper proposes the evaluation of SPIDAR-tablet's haptic accuracy. In the former earlier study, we had developed SPIDAR-tablet. In this paper, we made on experiment confirming that the haptics was presented correctly by the subjective questionnaires. As a result, we proved that the haptics SPIDAR-tablet presented is accurate.

View full abstract

We present a method for interactively retargeting "Manga" drawings. The method extracts important areas of the image using scribbles by the user. It then computes candidate images using cropping and seam carving. Finally, the user interactively combines these two effects to create the best result.

View full abstract

The recent development of the information society, there are many on the multimedia data on the Web. Especially in Japan, there are many comic images, but effective methods of searching for the classification category is not present. Need for a system that stores and manages image content is clearly increasing in the future. In this paper, we propose a system for classification category of Comic Images and evaluate it.

View full abstract

The purpose of this study is to enable partial image retrieval in the query-by-sketch image retrieval. We propose the segmentation method using edge-based feature extraction for query-by-sketch image retrieval. Experimental results of applying to 20,000 images in Corel Photo Gallery show that this method is effective.

View full abstract

This paper experimentally shows denoising performance of our previously proposed color-image denoising method with the anisotropic color total-variation semi-norm, and deals with a method for its automatic parameter setting.

View full abstract

This paper proposes a color-image wavelet denoising method with hard color-shrinkage utilizing inter-scale dependence of redundant wavelet coefficients. This paper experimentally demonstrates that our new method utilizing the inter-scale dependencies achieves a somewhat higher PSNR and removes wavelet-outlier artifacts more successfully than our previous color-image wavelet denoising method.

View full abstract

This paper presents a color-image denoising method with our previously proposed anisotropic color-TV semi-norm, which utilizes noise's signal-dependence for the denoising. Our newly proposed method can selectively remove signal-dependent noise more successfully that our previously proposed color-TV denoising method.

View full abstract

Extraction of important image regions is required for improving the quality of Image-Retargeting algorithm. We proposed the method to extract the region that is not appeared frequently (i.e., not repetitive) in the image as important region by exhaustive block matching. Our method extracts the important region with less number of parameters, and produces a better result compared with the conventional method.

View full abstract

In this paper, we propose the methods of depth estimation and free-viewpoint image synthesis using multiple sensors. The depth estimation method uses different block size at each pixel. And the method of free-viewpoint image synthesis distinguishes occlusion using ray vectors. The accuracy of image synthesis is improved by the proposed methods.

View full abstract

Previously we have proposed the hybrid redundant-DCT that is constructed by combining the redundant DCT and the shift-invariant Haar wavelet transform hierarchically, and demonstrated its good denoising-performance. However, this approach has the drawback that noticeable artifacts appear along sharp color edges. To remedy this drawback, this paper applies the Haar wavelet transform instead of the DCT locally to a block with an edge in it, and experimentally shows its improvements.

View full abstract