In this research, we generate a fluctuating moving image from one image. In the past research, a pseudo height was generated from an image, and shake the surface of that height and the luminance of the original image is changed using the bump mapping method to generate it. However, in this research, the fluctuating cycle is simple in the whole area, and the change in the saturation is not considered. Therefore, in this research, paying attention to the size of the object to be shaken, a moving image is generated by giving a plurality of cycles and shaking the brightness and saturation of the image.
In recent years, moiré type 3D display has been researched. In the conventional display device using moiré, the one with the depth is not the displayed object itself but the background of the display device. Therefore, the displayed object had a pseudo depth rather than the actual depth. Therefore, in this paper, we propose a moiré type 3D display device consisting of a LCD and a lenticular lens in order to realize a 3D display which can display the object with actual depth. In the experiment using the prototype device, it was confirmed that the image with actual depth can be displayed by moiré interference caused by the lenticular lens and the periodic pattern displayed on the LCD.
Regarding Depth-from-Focus methods，which estimate the depth from a set of images taken by the
different focus,a method，which uses the relation between the lumiuance value of a target pixel and the mean value
of the neighboring pixels，and the relation follows Poisson distribution，has been proposed. However，the mean value
is computed using different window sizes，so an enormous memory and computation time are rerequired. Additionally，
in the calculation，box filtelring to compute the mean is used twice，so a target pixel has as trong relationship with
the neighboring pixels，which makes it difficuIt.to perform pixel-by-pixel concurrent calculationfor redllcing the
calculation time. In this paper，to address these problems，we consider modifying the image-wise calculation to the
pixcl-wise calculation，and propose a method that replaces the calculation with one filtcring by approximating the
formula. This can reduce the memory consumption and allows us to use pixel-wise concurrent calculation.
In this report, we propose an image completion method considering texture representation. In the conventional method, there is a problem that an appropriate texture cannot be reproduced in some images which some texture strongly arrears. Therefore, in this research, as postprocessing of the conventional method, we use Poisson Image Editing to interpolate appropriate texture. However, when Poisson Image Editing is applied to an area including color boundary, some color blur occurs. Therefore, proposed method extracts color boundary automatically and applies Poisson Image Editing to only on area other than the boundary. At the time, by using Split-and-Merge Segmentation, we select appropriate texture for interpolating automatically. By this method, it enable to reproduce both color and texture with high precision.
The illumination enhancement is important for the computer vision applications using outdoor. Although many studies have been reported to solve this problem, they contain various problems. Therefore, we used a method combining variational retinex model and bright channel prior to suppress contrast over-emphasis and noise. We also reduced the computational cost by using FFT as a method to solve the quadratic optimization problem. Furthermore, in order to suppress blur artifacts, we used activation map changed to a form applicable to night images. In our experiments, we confirmed the effectiveness of our method, and improvement of detection accuracy by object detection algorithm.
More than one hundred million individuals around the world suffer from color vision deficiency (CVD). Several recoloring algorithms have been proposed for CVD compensation. To address the problem that the compensated images generated by existing dichromats assistance methods may lose effect on a part of the anomalous trichromats, this paper proposes a new recoloring algorithm adopting an optimization model adapted to the degree of CVD. Contrast enhancement and naturalness preservation are served as the goals of optimization model and represented as an energy function in the recoloring model. The unction is minimized to generate the compensated image which preserves the contrast and naturalness in the original image.
In recent years, image quality has been regarded as important by improving display technology. The most common method for improving image quality is image enhancement, and estimation of high frequency components is important for image enhancement, but it is important to consider color in the case of color images. In this research, I propose a resolution enhancement of color image using Laplacian pyramid. Conventional research assumes images of one channel, and there was a problem that color can not be preserved if the resolution of three channels of the color image is independently increased. In this research, I tried to increase the resolution in the HSV space, and improved this problem by using bicubic interpolation to expand saturation and hue. I also defined a new bicubic pyramid and examined its effectiveness.
Research that obtains 3D feature points with high repeatability from the 3D point cloud is useful for
improving the precision of the shape，or for reducing the calculation amount，an d thus it can be said to be an important
field. In recent years，it confirmed that the VKOP feature point extraction method，
which allocates feature points at
virtual positions utilizing the plane portion within the point cloud: is effective in scenes containing many artifacts. In
order to further improve this repeatability，methods have been proposed in which plane equations are pre-filtered or
feature points are assigned with the area of the plane as likelihood. In this paper，we proposed methods with likelihood
by adopting different and combining each methods，an d compared its performance.
If videos viewed by viewers have screen shake information，it is one of the important issue to prevent
viewers from VIMS (visually induced motion sickness). Then，a screen shake extraction method using histogram
of motion had been proposed to prevent viewers from screen shake，which was a factor of VIMS. However，the
effectiveness of this study was evaluated for only horizontal shake. In this study，the extraction rate of screen
shake is considered by using a temporal correlation determination method with motion vectors (MV) feature
information for horizontal，vertical，and 2D screen shake. In addition，the extraction accuracy by the number
of group in MV histogram is evaluated by the simulation experiments，and the effectiveness is considered.
A data hiding method using region division and histogram manipulation is proposed in this paper.
The original image is divided into 2 non-overlapping sets based on the regions defined by a binary image，where
4 different histograms are further obtained from the original image. Next histogram manipulation is applied
to insert data in the original image. In this paper，a post-process method is proposed to improve the degree of
reversibility，which improves the image quality of the reconstructed output image. Experiments were carried out
to verify the basic performances of the proposed method.
Recently, research of dehazing for improving the visibility of images has been conducted widely. There is a method based on Dark Channel Prior as removing haze from single image, and it is simple and is used widely. However, there is a problem that white objects and illuminations are wrongly recognized as atmosphere light. Therefore, in this research, we propose a method using Saliency Map in estimating atmosphere light. In our method, we prevent from estimating atmosphere light from white objects and illuminations because we estimate it from regions with low saliency. As a result, we create clearer image than previous method, and we show that proposed method is effective.
Gaze information about human is important to judge the attention function in the human. For example, gaze information during car driving and cooking are important. We suggested being able to evaluate attention function from activities of daily living. We acquire gaze information using an eye tracker. The eye tracker is able to estimate high accuracy, however, it is expensive and difficult to operate. In this paper, we discuss a method to estimate gaze without calibration. Specifically, we estimate gaze from visual attention model inhuman. The visual attention model is made from a bottom-up model and a top-down model. However, the brain in human has parts that react to face, body-parts and so. Therefore, we defined a category-specific attention, integrates previous model and category-specific attention, and estimate gaze information. As a result, we present that gaze is similar using the proposed model than the previous model.
As of this moment，it is argued that there are about 285 million patients with visual impairment in the world and about
60 million patients among them suffer from visual field disturbance. The prior approaches to assisting the patients with visual field
disturbance rely mostly on non-visual sensation. In contrast，a vision assistance method proposed in this research attempts to
superimpose judiciously processed surrounding scene onto the limited visual field of the patients. This article reports main results
of preliminary evaluation experiments on the proposed method. The design of sophisticated experiments focusing on the
characteristics of ameliorated visual field is also discussed.
Although liquid embolization is performed in overseas as a preventive surgical procedure for subarachnoid hemorrhage caused by rupture of a cerebral aneurysm, it is unauthorized in Japan for the sake of safety. Then, the authors developed droplet formation simulation using particle method for its safety verification. The embolic material used in liquid embolization is highly viscous compared to water, so that reducing the simulation time step is necessary according to von Neumann condition. Otherwise, it takes a lot of time to analyze. Therefore, in this study, the viscous term is implicitly solved and the simulation time is reduced. Finally a simulation has been performed, where an embolic material is injected into a water tank that mimics a cerebral aneurysm through a circular tube that imitates a catheter. In addition, we have confirmed that a droplet has been formed, which was similar to the experimental result.
To render a scene with fluorescence, spectral rendering should be used because fluorescence is a strongly wavelength dependent phenomenon. We propose a method for rendering fluorescence under global illumination environment efficiently by using importance sampling of wavelength considering both spectra of light sources and fluorescent materials.
Extracting metadata from contents is becoming more important in production support and Internet distribution. The task of extracting metadata by manual operation requires much time and effort, and it is impossible to analyze them in detail then and there. We developed a system that can automatically recognize and tag the team name of each runner in the video of the road race program with a high degree of accuracy in real time using image recognition technology.