After spending several years of research in the field of computer graphics and computer vision, I am currently working on image processing for synthetic aperture radar (SAR), especially exploring the possibility of fast image formation from the radar data and improving image quality using deep learning-based techniques.

We have developed a method for 3D measurement of human organs by endoscopic images. Since human organs are small, have limited space, and have similar surface patterns, it is difficult to reconstruct them from an ordinary method. To solve this problem, a small projector is inserted into the endoscope and a pattern is projected from the projector.The accurate correspondence between the projected pattern and the original pattern is the key to the measurement. Our method uses machine learning… Show more We have developed a method for 3D measurement of human organs by endoscopic images. Since human organs are small, have limited space, and have similar surface patterns, it is difficult to reconstruct them from an ordinary method. To solve this problem, a small projector is inserted into the endoscope and a pattern is projected from the projector.The accurate correspondence between the projected pattern and the original pattern is the key to the measurement. Our method uses machine learning to find the correspondence between them and achieves a robust and highly accurate shape measurement. Show less

We have developed a method to estimate the shape of transparent objects such as water droplets from the images. Besides water drops are transparent and their shapes change over time, it is difficult to measure their shape with ordinary equipment. To solve this problem, we developed a deep learning method that inference the shape and density of water drops as a displacement map from an image. The time-varying part of the distribution is also modeled by the thermal diffusion equation based on… Show more We have developed a method to estimate the shape of transparent objects such as water droplets from the images. Besides water drops are transparent and their shapes change over time, it is difficult to measure their shape with ordinary equipment. To solve this problem, we developed a deep learning method that inference the shape and density of water drops as a displacement map from an image. The time-varying part of the distribution is also modeled by the thermal diffusion equation based on our observation. Show less

We developed a realistic image display method that took into account human visual perception. In the real world, in dark to bright scenes, the lighting range that we encounter in daily life is vast. On the other hand, the amount of light that a display monitor can emit is limited. Therefore the gap between the situation that a video is taken and seeing it on a display monitor appears. In our research, we considered the phenomenon that the perceived color changes according to the… Show more We developed a realistic image display method that took into account human visual perception. In the real world, in dark to bright scenes, the lighting range that we encounter in daily life is vast. On the other hand, the amount of light that a display monitor can emit is limited. Therefore the gap between the situation that a video is taken and seeing it on a display monitor appears. In our research, we considered the phenomenon that the perceived color changes according to the brightness level of the environment, the method achieved to bring a real impression in the limited dynamic range of a display monitor. Show less