There are several ways to represent 3D scene information. One popular way is based on N-view plus N-depth representation. In applications based on this representation, efficient compression of both view and depth is important. In this paper, we present a depth coding method that uses depth downsampling and a novel depth upsample filter that uses the color view in the depth upsampling. Our method of depth down- and up-sampling is able to reconstruct clear object boundaries in the upsampled depth maps, and, therefore, we can obtain a better coding efficiency. Our experimental results show that the proposed depth resampling filter, used in combination with a standard state-of-the-art video encoder, can increase both the coding efficiency and rendering quality, particularly at lower bit rates.
We present a method of generating new view-point movies for baseball games. The most interesting view-point in a baseball game is looking from behind the catcher. However, because the view is interrupted by the umpire and catcher, setting up a camera behind the catcher is impossible, and setting up a camera on the catcher's head is also impossible. Therefore, we removed the obstacles (catcher and umpire) from the movie captured from behind the catcher, and generated a see-through movie by using multiple cameras to recover the pitcher's appearance. Our method consists of two processes; recovering the pitcher's appearance by homography, and detecting obstacles by using Graph Cut. To clarify how effective our method is, we generated a see-through movie by applying our method to multiple-camera movies taken in a baseball stadium. In the see-through-movie, the pitcher appears through the catcher and umpire.
The characteristics of line drawing video communication are described in this paper. In the experiment, an original video and a line drawing video were compared. Subjects performed the same task using both video systems and their performances were assessed in terms of the number of utterances in the conversation and the subject's switching pause. The subject's conversation strategy was extracted based on gaze measurement. In the case of the subujects who often gaze at a partner's eyes during conversation, the switching pause for the line drawing video was significantly longer than for the original video. In both video formats, subjects who does not gaze at a partner's eyes had more switching pauses and utterances than the subjects who do.
Tracking objects at a distance is a problem in wide area surveillance. This paper introduces a realtime active vision system that can track a moving object from 1 to 200 meters. This visual servo system is mainly structured with three mechanical freedoms: pan, tilt and zoom. An eagle eye mechanism is proposed to obtain a wide field of view and high resolution on the target. A fusion scheme has also been developed that combines the results of the three separate zoom cameras to detect and track remote targets. A high performance pan-tilt unit and an algorithm of high frequency motor control are proposed to achieve smooth tracking. The experiments of tracking a person walking outdoors and indoors have been carried out. The results show that this system is valid for tracking in a wide area.
In the current Super Hi-Vision (SHV) camera system, a camera operator has difficulty in finding the best focus because the resolution of SHV video is extremely high compared to that of the viewfinder (VF). Therefore, a video engineer adjusts the lens focus on a high-resolution monitor on a remote site. We propose a focus-aid signal that enables a low-resolution VF to be used to focus the SHV. The signal is generated from the high spatial frequency components of an SHV video and mixed with an VF video. We first discuss the difficulty of focusing the SHV camera on the basis of the MTF calculation including the lens defocus. We then simulated the focus-aid signal generation by using an SHV video and evaluated the property of the generated signal. We finally developed a prototype focus-aid signal generator and the results of our experiments show that the signal provided good focus in an SHV system.
Intellectual property, clearly, is a major component for any company's competitive advantage in the 21st century. We can see many industries where intellectual property played important roles in aiding competitive advantage in the past. However, recognizing IPR importance is not sufficient. For example, the Japanese electronics industry lost its leadership in the DRAM and DVD business despite having strong technology superiority in terms of IPR. This paper deals with patents that are important components of intellectual property; specifically, this paper tries to define the competitive advantage of technology utilizing the two-dimensional framework of IPR exclusiveness and product life cycle.
We present a reversed unsharp masking technique for sharpening images by producing hue halos around edges. This technique is used to binarize images. In the reversed unsharp masking, a detailed component in the usual unsharp masking is reversed and added to a base component. Images sharpened with this masking are then thresholded to generate non-photorealistic binary images.
A modified window bilateral filter has been developed. Its denoising capability is superior to that of the bilateral filter. In this letter, we describe how the modified window bilateral filter, called the segment bilateral filter, evalutaes the similarity between pixels with selected pixels belonging to the same segment as the target pixel. Experiments demonstrated that the segment bilateral filter preserves object contours and removes noises in images better than the window biltaeral filter.