A method has been developed for automatically classifying baseball video scenes into some events that describe their content.The baseball scenes are patternized using a set of rectangles with image features and motion vectors.The basic unit for patternization is a shot.For the second shot of each scene which includes significant information for event-classification,a partial shot generated by dividing the shot is used as a processing unit.The scenes used for training are expressed as sequenced symbols based on the patternized data for shots and partial shots.“Event-unknown”baseball scenes are assigned “event-indexes”(i.e.,homerun,single,walk,etc.) using discrete hidden Markov models that have been trained with the training symbol sequences for each kind of event.An experiment using videos of seven Major League Baseball games produced good results,demonstrating that this method can automatically classify events with high accuracy.

View full abstract

We have improved a previously reported method for extracting and judging camerawork for an online training-oriented video-shooting navigation system.It uses video grammar for evaluating user's high-speed camerawork to assist them in effectively shooting video that can be easily edited.High-speed processing is required in the navigation system,so we use a luminance-projection correlation method to extract the camerawork parameters at high speed.However,the accuracy is insufficient.To improve the accuracy,we combined the luminance-projection correlation method with a split tensor histogram method that classifies each frame into one of seven camerawork types.Camerawork analysis results showed that the accuracy was significantly improved.

View full abstract

We aim to develop a video-shooting learning system using video grammar and a real-time,on- and offline navigation system.As an initial step,we have developed an online navigation system that uses video grammar for evaluating users' high-speed camerawork to assist them with effectively shooting shots that can be easily edited.Previous camerawork analysis results showed that by classifying each frame into one of seven camerawork types,high-speed processing accuracy is improved.The system issues three types of alarms depending on the camerawork and the video grammar while a user is shooting that guide the user toward a specified shot.Users thus can naturally acquire a shooting style by trying to decrease the number of alarms and can obtain the specified shot without having to consider the video grammar.

View full abstract

Normal maps play an important role in realistic 3D image rendering.They are used to express pseudo roughness of the 3D objects' surface with a small amount of polygon data.In this paper,a fast and efficient normal map compression algorithm using vector quantization is developed based on a quality estimation model of the rendered images.Optimal encoding strategy is mathematically obtained from the quality estimation model.Experimental results demonstrated that the algorithm developed in this paper gives better result than other previous approaches especially in higher PSNR region.As a result,it has been made possible to encode normal maps considering the quality of the resultant rendered images without actually rendering the images.Encoding time is made reasonable by considering the limited signal space of normal maps and employing inner product in the nearest vector search. In addition,decoding time is also very small,which is also an important factor for practical usage.

View full abstract

[in Japanese]

View full abstract

We propose a framework that enables vision-based navigation systems to more robustly recognize signboards.It is designed mainly for signboards with uniform color regions,which can be strongly affected by occlusion and reflection.The framework comprises two stages:preprocessing for extracting uniform color regions and recognizing them.In the first stage,the cumulative frequency,morphology gradient,and Mahalanobis distance are used for the extraction.In the second stage,a new matching method for identifying templates and a new relative method for verifying colors are used.Testing of a system using this framework and 300 images taken under various conditions showed a recognition rate of 98%,demonstrating the framework's powerful ability to recognize signboards.

View full abstract

The relationship between the statistical properties of a color signal and the conditions for perceptual color transparency was investigated. First, we investigated the physical change in color statistics by overlapping transparent color filters. The transparent color filters shifted the average of color signal and reduced its variance. Moreover, the color and luminance signals were highly correlated. A transparent field was perceived in the simulated scenes which consist of color noise with similar statistical changes caused by transparent color filters. We also performed a psychophysical experiment to investigate what effect color statistics have on surface segregation. The results suggest that second order statistics (correlation) is used in the visual system as one of the cues for perceptual color transparency.

View full abstract

A color demosaicing method has been developed that effectively performs within local areas that are rich in high spatial frequency variations.Such areas are likely to cause strong artifacts after demosaicing because of the inherent difficulty in subsampling.In visually tolerable image restoration,the spatial correlation between different color components is exploited to consistently maintain favorable transitions in the component intensities.Missing primary color values on the Bayer pattern color filter array are estimated using an edge-directed interpolation of primary color difference signals.The local transitions of the primary color differences are then smoothed by spatially adaptive filtering.Experimental results showed objectively-good and visually-preferable performance in terms of PSNR,CIELAB color difference and visual appearance,particularly resulting in less false color.

View full abstract

We have investigated real-time calculations and display optics for a full-color full-parallax hologram.In a previous study,we created a full-color rainbow hologram that eliminates vertical parallax and thus reduces computational complexity.We have now investigated the use of the image hologram for better color reproduction and full-parallax reconstruction.The full-color hologram was displayed on a holographic television that uses a part of the original optics and also uses LCoS panels from a conventional video projector to separate and combine color components.Using the experimental results,we evaluated the reconstructed image of an image hologram and were able to represent the color reproduction,depth,and full parallax of the image.

View full abstract

We have developed a stereovision method that uses a monocular head-mounted display(mono-HMD). In the past, both images in stereovision have been set in either a virtual world or in the real world. In the former case, a person wearing special glasses can see 3D images, but other people can't even see 2D images. In the latter case, viewers can only see 3D images from a particular location. Our method solves both problems. In our method, one image is set in a virtual world displayed in a mono-HMD, and the other image is set in the real world. The viewer wears the mono-HMD and moves her body to align one retinal image with the other retinal image. The brain's vision center then perceives the two images as one stereoscopic image. Our experiments have shown that 3D images can be viewed stereoscopically using our method.

View full abstract

In this paper, we assess the media synchronization quality of preventive control schemes at media sources and media destinations for voice and video communications.Two preventive control techniques are considered at the sources:Advancement of transmission timing with network delay estimation and temporal resolution control of video.Three preventive control techniques are considered at the destinations:Change of buffering time with network delay estimation, preventive pausing,and preventive shortening of output duration.By experiment, we make a comparison of the media synchronization quality of preventive control schemes that use these preventive control techniques.

View full abstract

A fast atom searching method for matching pursuits in a high-efficiency video coding system is described in this paper. The immense amount of operations is needed for the atom searching in matching pursuits,so speed-up in the searching algorithm is indispensable.We propose an atom searching algorithm that is based on both a correlation between the high signal-energy regions and optimal matching points and the correlation between the highly efficient approximated points, and to improve the computational complexity by reducing the searching points.

View full abstract