Feature point extraction, which is one of the basic technologies of 3D shape information processing, is being studied by an approach of how to find points with high repeatability and distinctiveness. From the viewpoint of ensuring distinctiveness, the focus has been on searching for the points at the tip ofthe protruding shapes, but it is difficult to ensure the repeatability of such shapes. On the contrary, VKOP (Virtual Keypoint Of Polyhedron) focuses on the high repeatability of equations obtained from planar surfaces and locates feature points at virtual positions, has been confirmed that it has higher repeatability than the conventional method. On the other hand, VKOP strongly depends on the plane estimation method and its parameters. In this paper, we propose a method that aims to reduce the dependence on the plane estimation method and improve the repeatability of VKOP with indexing the likelihood of the estimated plane equation. For this likelihood, we use a new index calculated based on the thickness of planes caused by the characteristics of the sensor and the addition of random noise. From the experimental results, it was confirmed that the introduction of the likelihood improves the repeatability of VKOP and shows more than four times the repeatability in a shorter processing time than the conventional feature points.

3D Point cloud registration is the problem of integrating 3D point clouds obtained from multiple viewpoints into a single coordinate axis, and is often used as a preprocessing for modeling and object recognition. In general, highly accurate registration methods require a large computational complexity because the processing is proportional to the number of points in the source point cloud. In order to reduce the amount of computation, the keypoint patches extraction method has been proposed, which uses a point cloud extracted only from the points around the keypoints. However, since the number of points is small, there is a problem that the accuracy of registration will deteriorate if keypoint patches are extracted in nonoverlapping areas. Therefore, we propose a 3D point cloud registration method that adaptively changes the position of keypoint patches. The method extracts the extra keypoint patches in advance and selects only the overlapping parts in those. This operation was implemented by a genetic algorithm that optimizes the overlap parts using the tentative registration results of the selected patches. The experimental results show that the proposed method achieves more accurate registration for point cloud pairs with smaller overlap regions than the conventional method.

When a disaster such as a large earthquake occurs, there is a need to convey quickly and accurately disaster information to victims. On the other hand, for some of the Deaf, whose primary language is sign language, it has a large cognitive burden of information obtained via a text-based system, it is not easy to read in the emergency information quickly and accurately. Video communication system that can interact with sign language is required. However, it is difficult to transmit a video file of a large capacity lines that slow down the speed at disaster congestion or the like. In this study, we aim to establish a key-frame communication system. It is a technique to reduce the compression of the data volume by constructing a sign language video only key frames (frame images spatial features of sign language is out strongly). In this paper, for the sign language image generated based on the key frame communication system, it was evaluated experiment by the Deaf people.

The development of Convolutional Neural Networks (CNN) has produced remarkable great contribution to a wide range of image processing fields. On the other hand,there are premised on the existence of a large amount of annotated data. In addition to it, when a model learned in a domain is applied to a different domain, even if in the same task, there is no guarantee of its accuracy. This is a very important issue when deep learning and machine learning are applied in the field. If re-annotation re-learning is performed for data with such a domain difference, the same accuracy can be expected to be maintained again. But, semantic segmentation needs fine annotation and its high labor cost makes its application difficult. Histopathological image segmentation which is expected to drug discovery and medical image analysis is expensive due to its annotation cost, a wide variety of specimen and the need for the skills of histopathological experts. Therefore, the purpose of this research is to reduce the re-annotation cost of data in the new environment using the idea of domain adaptation. In this study, we focus on domain adaptation focusing on the class imbalance problem. The output of the class with few labels tends to be unstable in the class imbalanced data We proposed new cost function which is focusing on the class of few labels with histopathological image. As you can see our result, we achived imporvement of this problem. This made it possible to create a model that suppressed the class imbalance of the dataset in domain adaptation for pathological image segmentation.

Visual field defect (VFD) is a loss of part of normal field of vision, resulting in vision distortion or sensation of seeing through a narrow tube. VFD is difficult to recognize by most patients because of the filling-in mechanism in the human brain. Widely used perimeters such as Goldmann and Humphrey have serious technical limitations on the subjectivity of visual field and vision acuity assessment. In contrast, the active perimeter automatically assesses VFDs by analyzing the involuntary eye movement characteristics when searching for visual stimuli presented on the screen. However, this perimeter has issues such as patients with VFDs encounter problems on performing the eye-gaze calibration, visual field assessment covers up to 60 degrees, and physical burden due to head fixation during the testing. This study proposes a high-performance active perimeter based on a high-speed eye tracking system in a Head-Mounted Display (HMD). The proposed perimeter has several features such as not requiring fixation of the head, testing up to 90 degrees Field-of-View (FoV), accurate pupil extraction, one-point gaze calibration, and visibility judgment by saccade latency and saccade count. A successful visual field testing has been conducted by 10 visually healthy subjects to recognize 76 visual stimuli randomly presented within 90 degrees FoV. Each testing was completed in about 14 minutes per subject, confirming that it significantly reduces the physical burden on the subject.

The random phase-free method is expected to be applied to holographic projections because it can produce a larger reproduction image than a hologram without using random phase. The random phase-free method has been shown to be effective for amplitude holograms, but it has not work well for kinoforms. On the other hand, the down-sampling method, which produces good reproduced images in kinoforms, cannot produce an image larger than the hologram size. In this paper, we propose a method to improve the image quality of kinoforms by using the random phase-free method in combination with the down-sampling method to obtain large reproduced images. As a result of simulation experiments, it was confirmed that the image quality of reproduced images from kinoforms can be improved.

In this paper, we propose a method of numerically approximating vorticity of simulated fluid and extracting their various geometrical characteristics. Our method uses quadratic forms in the approximation process, which can determine the structure of quadratic geometry of vorticity (ellipse, parabola, hyperbola). If the result is an ellipse, we can further extract useful geometrical information such as the center and major/minor axis length of an ellipse. We also investigate the relationship between these geometrical features and physical features of vorticity and show that an ellipse is near the center of the vortex while the parabola or hyperbola is highly related to the flow strain. Our method can be used to improve geometrical artifacts from existing fluid simulation methods such as vorticity confinement.

While automation has advanced in various fields in recent years, the wrapping of souvenirs is often done manually. In recent years, automation in various fields has been progressing, but the wrapping of souvenirs and other gifts is often done by hand. Manually wrapping box-shaped gifts requires personal experience and skills to make them look good. In this study, we developed an application that automatically generates a wrapping paper design that maintains the continuity of the pattern at the end of wrapping. We assumed that the texture to be generated is a plain stripe, the object of wrapping is a rectangular box, and the technique is diagonal wrapping. In order to generate stripes, we first define one of the large surfaces of the box as the ”main surface” and generate stripes on the main surface. The pattern on the main surface is mirrored on the part of the wrapping paper that faces the main surface. Finally, the side portions are generated, and the wrapping paper generation is completed. We used the functions we created in practice. As a result, we were able to maintain the continuity of the pattern on the bottom and sides.