Automated surface anomaly detection is important in industry. The goal with this task is to classify the query image and its pixel as either “Normal” or “Anomaly” with high precision for industrial products. Unlike general two-class classification tasks, we can only obtain normal images (i.e., anomaly images cannot be obtained) for training because anomalies rarely occur in well-controlled production environments such as modern manufacturing factories. We propose a transfer-based anomaly detection method called Defect Representation Transfer-based Anomaly Detection (DRepT). This method can generate natural anomaly images for the target domain (i.e., real factory with only normal images) using knowledge of anomalies obtained in the source domain (i.e., known datasets with normal and anomaly images). We first focused on an anomaly image of an industrial product consisting of two elements: “background texture” and “defect representation”. At the source domain, a background texture image is generated from an anomaly image with texture invariance (i.e., defects are only removed) by using generative adversarial networks, and defect representation is modeled as the difference between an anomaly image and its background texture image with the newly proposed Gaussian mixture transmittance mask. At the target domain, anomaly images are generated by giving defect representation to normal images, and a anomaly detection model is finally trained. Compared with a method without real defects of other domains and a method based on pasting of the defect images, DRepT achieved highly precision.

In this study, we propose a video-based facial expression recognition method that utilizes the "emotion-wheel model." In research on emotions in the field of psychology, a model has been proposed in which basic emotions are arranged in a circular pattern, such as where "happiness" and "sadness" are opposites. Therefore, we utilized this knowledge as an inductive bias to improve accuracy by embedding features that are consistent with the emotion model in the process of class identification for recognizing facial expressions in videos. As a result of a performance evaluation using the CK+, MMI, and AFEW datasets, the recognition rates of the proposed method are 98.78%, 81.95%, and 55.31% for each dataset, which are 3.67%, 6.34%, and 4.9% higher than the baseline method, respectively. Furthermore, the proposed method outperforms the state-of-the-art method on MMI and AFEW.

It has been inferred that the Philippine fault on Leyte Island has been creeping. On 6 July 2017, an Mw 6.5 earthquake struck northern Leyte, suggesting that not all parts along the fault have been creeping. We conducted InSAR time-series analyses using ALOS/PALSAR data (2006-2011) to estimate the spatial variation of the fault creep rate, and fault slip modeling of the 2017 earthquake using the displacements obtained from ALOS-2/PALSAR-2 interferometry. We found that the section ruptured by the 2017 earthquake had been locked (lack of fault creep) while the neighboring sections, in northernmost and central areas of the island, have been creeping with rates of up to 30 mm/year. We speculate from the creep analysis that the size of the 2017 earthquake would be the maximum-class earthquake along the Philippine fault in the northern half of Leyte. Considering the similarity in size and location, we infer that the 1947 earthquake also ruptured the same locked portion as the 2017 earthquake. Our findings demonstrate the usefulness of InSAR time-series analysis to identify future earthquake-prone sections along active faults.

We present postseismic deformation following the Kumamoto, Japan, earthquake sequence in April, 2016 detected with InSAR during ~2 years. We process ALOS- 2/PALSAR-2 and Sentinel-1 SAR images that were acquired right after the mainshock. We detected concentrated deformations along the source faults, Futagawa and Hinagu faults, as well as off-fault deformations. We found LOS displacement consistent with right lateral motion of Futagawa and Hinagu faults, which implies afterslip of causative faults. Displacements locally exceed 10 cm during 2 years. We observe various patterns of temporal evolution of LOC changes. LOS decayed with time constant of about 200 days in a part, while LOS changes accelerated in summer of 2016 in other area. LOS changes also appear to correlate to geology.

We propose a fast and accurate 6D object pose estimation from a RGB-D image. Our proposed method is template matching based and consists of three main technical components, PCOF-MOD (multimodal PCOF), balanced pose tree (BPT) and optimum memory rearrangement for a coarse-to-fine search. Our model templates on densely sampled viewpoints and PCOF-MOD which explicitly handles a certain range of 3D object pose improve the robustness against background clutters. BPT which is an efficient tree-based data structures for a large number of templates and template matching on rearranged feature maps where nearby features are linearly aligned accelerate the pose estimation. The experimental evaluation on tabletop and bin-picking dataset showed that our method achieved higher accuracy and faster speed in comparison with state-of-the-art techniques including recent CNN based approaches. Moreover, our model templates can be trained solely from 3D CAD in a few minutes and the pose estimation run in near real-time (23 fps) on CPU. These features are suitable for any real applications.

Object detection has become an indispensable part of video processing and current background models are sensitive to background changes. In this paper, we propose a novel background model using an algorithm called Co-occurrence Pixel-block Pairs (CPB) against background changes, such as illumination changes and background motion. We utilize the co-occurrence “pixel to block” structure to extract the spatial-temporal information of each pixel to build background model, and then employ an efficient evaluation strategy to identify the current state of each pixel, which is named as correlation dependent decision function. Furthermore, we also introduce a Hypothesis on Degradation Modification (HoD) into CPB structure to reinforce the robustness of CPB. Experimental results obtained from the dataset of the PETS 2001, AIST-Indoor, SBMnet and CDW-2012 databases show that our models can detect objects robustly in strong background changes.

We have studied ground deformation in the Osaka and Kanto plains, Japan, using ALOS/PALSAR and revealed unique deformations related to active faults. In order to clarify their evolution after 2011, we analyzed strip-map mode images of ALOS-2/PALSAR-2 acquired during 2014-2015. In Osaka, we did not find deformation observed before 2011 such as uplift in southern Kyoto and subsidence along the Arima-Takatsuki fault, even after we corrected altitude dependent error of phase for interferograms with quadratic functions. We found less than 1 cm/yr LOS changes, but correlation with deformation is poor possibly due to short interval. We also found anomalous LOS decrease in a basin in northwestern Kanto in summer, which might be attributed to dense water vapor concentration.

In this paper, we propose a high-speed 3-D object recognition method using new feature values. Features for the object recognition method proposed in this study consist of three values. One is the Difference of Normals (DoN) feature value that has been proposed by Ioannou. The other two represent information about curvature. We use these three-dimensional features to recognize the position and pose of multiple objects stacked randomly. Because they are low-dimensional, high-speed matching can be achieved. We have also reduced the computing time needed for data matching by using only effective points selected on the basis of their estimated distinctiveness. Experimental results using actual scenes have demonstrated that the computing time is about 93 times faster than that of the conventional SHOT method. Furthermore, the proposed method achieves a 98.2% recognition rate, which is 17.9% higher than that of the SHOT method. Also, we confirmed that the proposed method achieves higher-speed matching and higher recognition success rate than the conventional methods.

We propose a reliable 3D position and pose recognition method for complicated scenes including randomly stacked objects. Conventional methods use a small number of features selected by analyzing a target object model for recognition. The small number contributes to high-speed recognition, but actually the features include both "true" and "false" features. True features exist only in the target object and not in other parts, so they are valid for correct recognition purposes. On the other hand, false features exist in both the target object and in other parts, such as contacting areas caused by multiple objects. As a result of their matching incorrect parts, misrecognition may occur. To solve this problem, we propose a new method that uses effective features selected by analyzing not only the target object but also contacting areas caused by multiple objects. For predicting contacting areas, we generated very real input scenes by using 3D Computer Graphics (3D-CG) techniques and a physics-based simulator. Features that have high discrimination performance in the feature space are selected and used for the matching process. The method is robust to disturbances such as feature variability and achieves high feature separability; these enable it to achieve good discrimination performance. The method achieves reliable and fast object recognition by using a small number of effective features that have high discrimination performance. Experimental results show that the method's recognition success rate is from 33.6% to 92.9% higher than that of the Vector Pair Matching (VPM) method proposed by Akizuki et al. and that its processing time is within 1.46 seconds.

We propose a fast and reliable 3D object detection method that can be applied for complicated scenes consisting of randomly stacked objects. The proposed method uses "3D vector pair" that has a common start point and different end points and it has surface normal distribution as the feature descriptor. By considering the observability of vector pairs, the proposed method has been achieved high recognition performance. Observability factor of the vector pair is calculated by simulating the visible state of the vector pair from various viewpoints. By integrating the observability factor and the distinctiveness factor proposed in our previous work, vector pairs that have effectiveness for matching are extracted and these are used for object pose estimation. Experiments have confirmed that the proposed method increases the recognition success rate from 45.8% to 93.1%, in comparison with the state-of-the-arts method. The processing time of the proposed method is fast enough for the robotic bin-picking.

An illumination-invariant background model for detecting objects in dynamic scenes is proposed. It is robust in the cases of sudden illumination fluctuation as well as burst moving background. Unlike previous works, it distinguishes objects from a dynamic background using co-occurrence character between a target pixel and its supporting pixels in the form of multiple pixel pairs. Experiments used several challenging datasets that proved the robust performance of object detection in various environments.

In this paper, we propose a practical object recognition system which consists of two functional modules. The first is object extraction module using a range image, and the second is a precise position measurement module using a grayscale intensity image. Both high-reliability and high-accuracy can be achieved by effective image integration. We also propose an idea of stereo vision with random-dot pattern projection as an effective way to obtain a range image.

This paper presents our real-time vision system for robots. The system employs a new internal image representation, in which the scene is encoded into three intensity level images. This representation is generated by Laplacian-Gaussian filtering followed by dual-thresholding. We refer to this image as three-level broad-edge representation. It reduces the computational cost of the normalized cross correlation. Thus the system performs the searching visual task as well as the object tracking task, which is not reliably achieved by the mean absolute error based correlators. Our prototype system achieves 32/spl times/32 pixel normalized cross correlation based matching from the 128/spl times/128 pixel search area in 2 ms under 9 MHz pixel clock image processor. This speed is fast enough for searching and tracking a single object at video frame rate.

Genetic labeling is a new labeling algorithm using Genetic Algorithm (GA). Although several applications of GA for low-level image processing such as line detection have been studied, they still require much computing time. We apply GA to the labeling for scene interpretation. The chromosome coding method we proposed is such that each bit represents the existence of an object. Genetic operation enables efficient labeling based on the building block hypothesis. We have developed a vision system for depalletizing robot using this technique. Object candidates are properly labeled, and the position of cartons is recognized. Through real image experiments, we estimated that genetic labeling is about 100 times faster than an improved enumerating method. Also we have proven that the reliability and speed of this system is practical.>