There is an increasing demand for high quality real-time graphics nowadays. Shadows play an important role to the realism of computer-generated images by enhancing depth perception and curvature reception. Due to their global nature, shadows introduce overwhelming complexity to rendering algorithms. Recently, screen-space ambient occlusion techniques started to flourish, and are now the de facto standard for real-time dynamic shadow synthesis. A few issues remain, though, such as the sampling quality and noise artifacts. The contributions of this work are two-folded: a variation of screen-space ambient occlusion that uses Summed-Area Tables, yielding satisfactory results yet performing better than previous attempts, and serving as a new application to the arsenal of Summed-Area Tables.

View full abstract

We present a new technique for artistic style transferring by considering the perceptual structure matching between source image and target images. Given a photograph, our technique first searches for an existing art work on a similar scene through using the computational model of GIST perception and then transfers both the color and brush texture from it by considering the structure matching between two images. The proposed method is fully automatic and can be used for transferring any styles. Subject studies have been conducted to validate the effectiveness of our newly designed GIST feature vector as well as the enhanced color and texture transferring considering structure matching.

View full abstract

In order to realize a safe and secure automotive society, many studies have been made on the ITS. In this paper, we propose a new method for detecting hazardous road surface conditions by using an omni-directional camera and polarization property, and the especially imaging unit. It is capable of simultaneous acquisition of parallel and vertical polarization in all directions. Real and mirror images, by the mirror was placed at the top of the unit, passes through the filter of parallel and vertical polarization, respectively. Omni-directional image is expanded into panoramic image. Parallel and vertical polarization components are extracted from the panoramic image. The amount of feature is calculated from the parallel and vertical polarization components. State of the road surface is identified from the feature.

View full abstract

In this paper, in order to maintain precision of detection of moving object, we propose novel multi-templates method based on orientation code matching under environment where multi-moving objects exist and where the change of illumination occurs. Further, by using an extended Kalman filter, we treat position estimation of moving object to obtain highly precise detection of moving object. In fact, we introduce an evaluation function defined by the distance of between information of detection position of moving object given by the past orientation code data and information of estimation position of moving object calculated by an extended Kalman filter, and then according to the evaluation value we decide the position of moving object on the image. By repeating the detection method above in time series image we track moving object. Finally, we perform experiment of tracking of mosquitoes which are arthropod, and then show effectiveness of our proposed method.

View full abstract

When producing the animation of a body motion from the dance notation, the dance knowledge is a key for accomplishing high-quality movement. This knowledge enables the dancer to know how to perform the correct movement from a movement notation score. This paper presents an approach for automatically simulating a CG animation from Labanotation scores. We achieve this goal by the integration of a CG animation with a dance-style interpretation module which is called an autonomous dance avatar. In our experiment, we implemented an autonomous dance avatar to perform Japanese stylized traditional dance Noh plays. The result shows that the autonomous dance avatar can reproduce Noh play satisfactory from Labanotation after it has been trained with the style of Noh play.

View full abstract

Biometrics has risk of forgery and theft. And a lot of biometrics cannot change the key that is used for personal verification at these cases. In this paper, we propose the multimodal verification system using hand-shape and signature. The signature is acquired as a trajectory of fingertip under Web camera, and the hand-shape is acquired at the same time. The whole verification system consists of hand-shape verification and signature verification. The system determines the input as genuine if it is verified as genuine by both verifications. Experimental results in this verification system have shown the effectiveness of the proposal system. It was confirmed that FAR was reduced from 7.8% to 5% in becoming false rejection rate FRR=0% by the multimodal.

View full abstract

The error diffusion method is used well for two-level dithered picture because of its high-quality processing compared with other dithering methods. It can be expanded to multi-level processing. However, the error diffusion method using multi-level processing causes significant deterioration of picture quality, when an input pixel level is near to a quantization level. In this paper, we propose a new error diffusion method to solve the above problem, which technique is to set the diffused error value with an upper limit level under the condition of the diffused error increase. The proposed method is found to be effective for image quality. In addition, we show that usual two-level error diffusion methods have a characteristic of edge emphasis.

View full abstract

In a contrast improvement, the high emphasis effect is important. However, if the emphasis effect is high, there is a problem that a noise occurs. The purpose of this paper is reducing noises by using the features of an image, and the paper proposes a new contrast improvement method that can control occurring noises. Specifically, the noise is identified by use of a first differential value and a quadratic differential value. A smoothing filter is applied to only the noises. The contrast improvement by the differential gray-levels histogram is applied. The experimental results showed that the resultant images with visually good and natural contrast were generated by the proposed method.

View full abstract

This paper proposes a high-speed match method by selecting the effective feature pixel in the template. Since the template matching has the disadvantage that needs long processing time, many high-speed image matching methods using information on the edge are studied. However, the edge of the image becomes unstable depending on the recording conditions. On the other hand, the change of the density value of the pixel around the edge is smaller than that of the pixel of the edge. However, its method doubles the number of pixels that should be examined. Therefore, the proposal method evenly divides the template image, then selects the pixel with the highest density value and the lowest density value in each division area. As a result, the proposed method had the same detection performance as ZNCC(Zero-mean Normalized Cross-Correlation) and SIFT(Scale-Invariant Feature Transform). Moreover, the proposed method was able to detect faster than conventional method.

View full abstract

Moving keyboard for eye-based Human-Computer Interaction: HCI is proposed for disabled and elderly persons, in particular. The current problems on eye-based HCI, (1) poor key-in success rate, and (2) slow key-in speed are to be overcome with the proposed moving keyboard. The keyboard is moved by the gaze of user looking at one of the enlarged semi-transparent keys located at the center and surrounding up/down/right/left hand sides. Selected key is always shown at the display center. Thus the selected key is easy to find and can be determined by blinking or by fixing one's eye on the key for a second. Experimental results of relations between key-in success rate and (1) variability of gaze location, (2) key size, (3) the number of keys, (4) camera resolution are introduced which show the validity of the proposed method. The detailed proposed system is described followed by performance evaluations to overcome the aforementioned problems.

View full abstract