In H.264 coding technology, macroblock coding mode is selected from much more candidates compared to conventional coding standards, and an appropriate mode decision introduces the successful coding performance. As for this adaptive mode selection, a simple design might degrade the coding efficiency. For instance, in an interframe coded picture, frequent selection of intra prediction modes independent on the image feature might result in degradation of subjective picture quality especially for the sequence with a global high speed motion. From this perspective, we propose an adaptive control for exploiting intra prediction mode within an interframe coded picture in order to achieve improvement of subjective picture quality while maintaining the same bit-rate.

View full abstract

In this paper, we propose a variant of mean shift filter (MSF), and we call it dynamic MSF (DMSF). In contrast to the original MSF, in which the set of samples is stationary, the proposed DMSF dynamically updates the sample set. Experimental results show that the DMSF converges to similar solution to that of MSF with smaller number of iterations than that of MSF.

View full abstract

We have developed a multi-lens camera system. This system can perform motion detection using one frame captured by a CMOS image sensor. Proper preprocess and comparison of four unit images enable us to get motion vectors accurate enough. We carry out experiments and confirm the effect of the proposed method.

View full abstract

This paper addresses shot classification and scene segmentation methods for MPEG compressed films. In addition, we propose a film recommendation framework according to features vector derived from the classification results. Proposed framework enables to recommend films without a priori knowledge and/or user ratings for the films.

View full abstract

We have analyzed the correlation in intra prediction mode of H.264 between low and high resolution images, which can be used for block classification for fast image super-resolution method. As a result, it can be expected that Intra16x16 mode can mostly be kept by super-resolution as well as flatness of corresponding blocks.

View full abstract

We previously proposed an efficient lossless coding scheme for color images. In this scheme, multiple predictors which exploit both spatial and spectral correlations are optimized for each image. This paper describes a method for determining the appropriate disposition of reference pels for each predictor.

View full abstract

This paper proposes an efficient H.264 bit rate converting method aiming to reduce the file size where the reconstructed video quality is improved as compared to the existing methods. This is achieved by optimizing the re-quantization parameters based on the original quantization parameters and macroblock bit amounts estimated from the input bitstream.

View full abstract

In this paper, we propose a robust watermark detection method for MPEG video up-converting in condition that the embedding method is not changed from our conventional method. Simulation results show that a watermark can be detected almost 100% from up-converted video by a small number of samples.

View full abstract

The information embedding by data structure of the color space of a color image was proposed. In the previous work, it is shown that the lifting of RGB color space contributes to the improvement of the quality of an embedding image. In this paper, a SN ratio and a decoding rate about embedding by 96th lifting are shown. And the suitable number of dimensions to embed for RGB image is also considered from a viewpoint of an embedded quantity and processing time.

View full abstract

We have proposed a DCT/DST hybrid video coding with the aim of reducing block distortion in block-based coding. This method employs an intra prediction using interpolation based on Inverse-DCT/DST. In this paper, we introduce DCT/DST coefficients filtering to improve the video quality.

View full abstract

Gaussian distribution is observed in vast mathematical/engineering areas. Contrarily to what happens for geometric and double-sided geometric distributions, there is no simple, instantaneous code for the normal distribution. This paper tackles this problem by converting Gaussian distribution into geometric distribution before applying Golomb codes.

View full abstract

The sea bream is being cultivated in the Regional Fisheries Research Laboratories, Kinki University. A culture sea bream occurs abnormalities in form owing to the imbalanced nutrition etc. Now, an expert presumes visually that abnormalities are normal for discernment of a sea bream fry, and an alternative is identified. The automatic discernment is indispensable in order to carry out reduction of cost etc. This research examines how to identify a normal sea bream fry and an unusual sea bream fry.

View full abstract

Kinki University is cultivating sea bream fry. Abnormality might occur in a sea bream fry's shape. A sea bream fry with a normal shape or an abnormal shape are being sorted out by the hand work now. This research aims at the automatic sorting by the image data processing to achieve the reduction in the cost and time.

View full abstract

This paper presents an automatic gender discrimination system, which can be applied for the marketing research. In general, the difference of appearance between man and woman can be used to distinct their sex. Several effective evaluating methods for gender discrimination are examined.

View full abstract

A method for setting specific areas on a face to extract feature quantities for the recognition of facial expressions is presented. In the previous method, the setting of the specific areas on a face was not occasionally achieved because the positions of facial parts are different from person to person. In this paper, a new algorithm for setting specific areas which can be applied even for the angled face is presented.

View full abstract

To achieve high recognition accuracy, feature quantities must be extracted precisely from the specific area on a face. To specify the particular area precisely on a face, facial inclination angle must be detected in advance. Based on the features of a face, such as eyes, mouth, the detection of the inclination angle of a face is executed by proposed algorithm.

View full abstract

The purpose of this study is to detect edges for the sketch image retrieval system. In this method, edges are detected with Canny edge detection to multiple-resolution images of database images. Experimental results of applying to 6,500 images in Corel photo gallery show that this method is effective in retrieving images

View full abstract

We proposed a presentation system for comparison of the medical images on neurosurgical planning. This system indicated the corresponding points of MRA and MRI clearly.

View full abstract

In the game analysis of badminton, it is important to acquire detailed data correctly. In this research, development of a video system and its image processing to support such game analysis is performed. In this paper, the image processing for extracting a trunk from a game video is proposed. In badminton, it is important in analysis that the information on a trunk is acquired, since the speed of a shuttle and the kinds of shot are influenced greatly. An experiment shows the result of having applied the proposed method to the video image of the international first class match.

View full abstract

This research proposed new methods of motion analysis based on classification of movement curves and motion tracking using 3-dimensional Hough transformation for time-sequence chest MR images.

View full abstract

This paper proposes a method for correcting videos which include flickers by flashlight. Frames are interpolated considering motion information between frames before and after flashlight. Experimental results show that the frames are appropriately corrected retaining continuity in camera motion.

View full abstract

This paper presents our system which remove only a projected image from a captured image. This system has been developed, and has improved the performance of the image separation more than the previous system.

View full abstract

To realize high-speed reading of videotapes, we propose an optical reading method of videotapes using magnetic garnet film. By putting a magnetic garnet film close to the videotape, recorded patterns are transferred to the film. The trial magnetic garnet film has large Faraday coefficient and recorded patterns of videotape are clearly transferred to the film.

View full abstract

2.5mm-thick 1inch hard disk drive with a recording capacity of 10 GB was demonstrated using a perpendicular magnetic recording medium with a soft-magnetic underlayer (SUL) of 50nm thick. The decrease of SUL thickness leads to improvement of bit error rate, attaining an areal recording density of 200 Gbit/inch^2.

View full abstract

In order to reduce the critical current density J_<c0> of current induced magnetization reversal, CPP-GMR devises with heusler alloys were fabricated. J_<c0> of CPP-GMR devise with Co_2MnGe is 1.6×10^7 [A/cm^2] and J_<c0> of CPP-GMR devise with Co_<75>Fe_<25> is 2.2×10^7 [A/cm^2]. Using heusler alloys, we successfully achieved the reduction of the critical current density.

View full abstract

A data page with moire which is interference between reconstructed beam and reflection beam in a CCD cannot represent high SNR. By compositing two data pages which have reverse moire patterns with each other, it is possible to cancel moire patterns and SNR can be improved.

View full abstract

Read/write characteristics of hologram multiplexed recording in photopolymer media are studied by angular multiplexed recording experiments and simulation using a nonlocal diffusion model.

View full abstract

Bit error characteristics of hologram multiplexed recording are studied from the view point of frequency bandwidth at angle-multiplexed recording.

View full abstract

We estimated that the luminescence properties of plasma display panel with a nano-size grain green phosphor. The luminous efficiency was on an equality with the conventional size phosphor coated panels, and the dispersion of minimum sustained voltage was drastically alleviated.

View full abstract

We have developed demountable evaluation equipment for field emitter arrays (FEAs) for ultrahigh-definition FEDs. Utilizing this equipment, we can easily evaluate properties of 0.15 mm pixel-pitch FEAs including focusing characteristics.

View full abstract

An organic light-emitting diode (OLED) panel driven by organic thin film transistors (OTFTs) has been fabricated on a plastic film. Highly efficient phosphorescent material was used for the OLED. We successfully confirmed active-matrix operation of the fabricated 16x16 pixels panel. The result suggests that the combination of OLEDs and OTFTs is promising to realize a flexible display.

View full abstract

We present the technique for reducing blur caused by out-of-focus of projector. We use a camera for estimating the Point-Spread-Function (PSF) in the image projected onto the screen. According to the estimated PSF, the original image is pre-corrected, so that the projected image can be deblurred. Experimental results show this technique can reduce out-of-focus projector blur.

View full abstract

A highly sensitive compact image sensor consisting of a 15-μm-thick HARP target and a Spindt-type field emitter array (FEA) equipped with an active drive circuit was fabricated for SDTV cameras. The number of pixels in the prototype sensor is 640×480 and the pixel size is 20×20 μm. Our experiments revealed that the prototype sensor could obtain sufficient resolution as an SDTV image sensor, high sensitivity and wide dynamic range by convenient operation.

View full abstract

We study the method of Next Generation Traffic Information Transmission by Digital Broadcasting System and develop the verification system of the method. We conclude that the method is available for the embedded systems.

View full abstract

Fundamental experiments on measurement of the gaze direction of a driver using Eyemark Recorder were performed. It was shown from the experiments that the gaze direction of a driver during driving concentrated in front and moved almost little right and left when the driver used a cellular phone.

View full abstract

The reflective characteristics of human eye and skin have the typical properties in the region of near-infrared wavelength. By using the characteristics, we have developed a camera system that can detect locations of eyes and skin from one-shot image. The camera system consists of a Multi-Lens optical system and a CMOS image sensor.

View full abstract

We evaluated realness of real objects and images at various resolutions. Both real objects and images are viewed through a synopter, which removes horizontal disparity and presents the same image to both eyes. Sixty-five subjects evaluate the realness of three real objects and their images at six different resolutions ranging from 19.5 cpd to 156 cpd. We found that realness of image increases as the image resolution increases up to the discrimination threshold based on visual acuity and remains constant above the discrimination threshold.

View full abstract

The well-known wagon wheel effect is analyzed by spatio-temporal sampling. This leads to establishing analysis methods for similar phenomena in moving pictures, such as apparent motion.

View full abstract

We measured the average MAF(Minimum Audible Field) by age for people from their twenties to eighties and investigated to adapt speech signals for elderly listeners based on their aural characteristics.

View full abstract

This paper discusses a method to generate baseball digest video using baseball ontology and scorebook. The methods extracts highlight scenes reflecting game movement and video editor's know-how.

View full abstract

In vein certification systems, the difference of distance between a hand and a camera, translation gaps, and rotation gaps are problems. Vein images are encoded using log-polar mapping to correct these gaps. We evaluate matching error in the template matching, and estimate the threshold for certification.

View full abstract

According to a report, human beings compare a resemblance between faces not wholly but partially. In order to calculate the similarity of human-face images, we proposed the novel method using the factor with which human detects resemblance in human faces. This paper evaluates the method by the subjective experiments with 10 persons and shows the effectiveness of the method.

View full abstract

A personal identification system based on the chain coded isodensity maps is presented. Isodensity maps can be converted to the chain codes. By using the coded isodensity lines, the improvement of the processing performance can be expected in terms of the processing time and the memory size required.

View full abstract

This paper presents a method constructing isodensity maps used for personal authentication system. In the previous method, the isodensity lines positioned near outline of a face was not extracted precisely the enough. Therefore, isodensity maps which include more individual features of a face for personal authentication system are examined.

View full abstract