ABSTRACT Unsupervised anomaly localization aims to identify anomalous regions that deviate from normal sample patterns. Most recent methods perform feature matching or reconstruction for the target sample with pre‐trained deep neural networks. However, they still struggle to address challenging anomalies because the deep embeddings stored in the memory bank can be less powerful and informative. Specifically, prior methods often overly rely on the finite resources stored in the memory bank, which leads to low robustness to unseen targets. In this paper, we propose a novel subspace‐guided feature reconstruction framework to pursue adaptive feature approximation for anomaly localization. It first learns to construct low‐dimensional subspaces from the given nominal samples, and then learns to reconstruct the given deep target embedding by linearly combining the subspace basis vectors using the self‐expressive model. Our core is that, despite the limited resources in the memory bank, the out‐of‐bank features can be alternatively “mimicked” to adaptively model the target. Moreover, we propose a sampling method that leverages the sparsity of subspaces and allows the feature reconstruction to depend only on a small resource subset, contributing to less memory overhead. Extensive experiments on three benchmark datasets demonstrate that our approach generally achieves state‐of‐the‐art anomaly localization performance.

Subspace clustering methods which embrace a self-expressive model that represents each data point as a linear combination of other data points in the dataset provide powerful unsupervised learning techniques. However, when dealing with large datasets, representation of each data point by referring to all data points via a dictionary suffers from high computational complexity. To alleviate this issue, we introduce a parallelizable multi-subset based self-expressive model (PMS) which represents each data point by combining multiple subsets, with each consisting of only a small proportion of the samples. The adoption of PMS in subspace clustering (PMSSC) leads to computational advantages because the optimization problems decomposed over each subset are small, and can be solved efficiently in parallel. Furthermore, PMSSC is able to combine multiple self-expressive coefficient vectors obtained from subsets, which contributes to an improvement in self-expressiveness. Extensive experiments on synthetic and real-world datasets show the efficiency and effectiveness of our approach in comparison to other methods.

Abstract This paper presents a novel multi‐scale template matching method that can be applied in unconstrained environments. The key component behind this is a general similarity measure is referred to as the diversity similarity measure against scaling, rotation, and illumination (DS‐SRI). Specifically, DS‐SRI exploits bidirectional diversity calculated from the nearest neighbour matches between two sets of points. Scaling and rotation changes are taken into consideration by introducing normalisation term on the scale change, and geometric consistency term with respect to the polar coordinate system. Moreover, in order to deal with the illumination change and further deformation, illumination‐corrected local appearance and rank information are jointly exploited during the nearest neighbour search. All the features of DS‐SRI are statistically assessed, and the extensive visual and quantitative results on both synthetic and real‐world data show that DS‐SRI can significantly outperform state‐of‐the‐art methods.

Horizon detection is useful in maritime image processing for various purposes, such as spatial orientation estimation of ship camera and detection of the significant region for post-processing. This paper proposed a novel realtime optimization-based method for detecting the horizon line in maritime images. Traditional methods defined the horizon line detection problem aim to detect the horizon line which perfectly divides the whole image into two regions as the sky and sea. Thus, the complication of traditional methods is the statistical distance metrics of distributions are calculated in the whole image by all combinations of the horizon line parameters. Moreover, traditional methods do not provide realtime processing to detect the horizon line from high-resolution images. To achieve real-time processing on high-resolution images, this study defines the local features of the horizon line using a vanishing line characteristic and the optimization criteria. In addition, the optimization process is improved by combining a genetic algorithm and a coarse-to-fine approach. Our method applied the Singapore marine dataset and the Buoy dataset. To verify the proposed method, our result is compared to the state of the art methods. We confirm the proposed method can accurately detect the horizon line under different scenarios in real-time.

Recently, deep convolutional neural networks (CNN) have become a new standard in many machine learning applications not only in image but also in audio processing. However, most of the studies only explore a single type of training data. In this paper, we present a study on classifying bird species by combining deep neural features of both visual and audio data using kernel-based fusion method. Specifically, we extract deep neural features based on the activation values of an inner layer of CNN. We combine these features by multiple kernel learning (MKL) to perform the final classification. In the experiment, we train and evaluate our method on a CUB-200-2011 standard data set combined with our originally collected audio data set with respect to 200 bird species (classes). The experimental results indicate that our CNN+MKL method which utilizes the combination of both categories of data outperforms single-modality methods, some simple kernel combination methods, and the conventional early fusion method.

This paper addresses the modeling and measurement of a small boat. In some fishing tasks, anchorage is not applicable in order to capture shellfishes or fishes efficiently. Currently, many fishermen are manually stabilizing boats simultaneously with the capturing task. We propose a boat modeling method named FAD learning. In this method, the free dynamics acceleration and actuator acceleration of a boat are learned using the online learning of two dynamics learning trees (DLTs), which are developed by us. In order to measure the position, velocity, and acceleration, we developed an image processing method with an underwater camera. In the experiment, the motor babbling of a boat was performed on a water pool. The dynamical data from the boat was learned by DLTs. The effectiveness of the modeling was confirmed through the validation of the velocity that was predicted by DLTs.

In this paper, we propose a pose estimation method of an object in three-dimensional space without the pre-learning for AR application in the mobile device. One of the implementation methods of augmented reality is using a two dimensional marker. A two-dimensional marker is not robust in the case of change in complicated three-dimensional rotation. In our work, this problem is solved by a three-dimensional marker using a cube. In the study of a pose estimation of the three-dimensional object, a model-based method using evolutionary method has been proposed. This method cannot estimate all poses, because the rotation angle of the target object is limited. In this paper, we propose the pose estimation method of a cube using a property of cube with evolutionary method. This method is possible to estimate all pose by using property related to rotation of the cube. We evaluated the proposed method by experiments using the static image of object in real-world.

Summary form only given. To make Virtual Reality (VR) world closer to reality, it is necessary to acquire much information from the real-world. Especially for simulation of transportation, it is important for VR world to obtain traffic flow in reality. This paper describes an application to compute traffic flow through a sequence of images captured with video camera. In this search, camera is not set above the roadway but beside the roadway.

In this paper , we propose a new statistical learning algorithm. This study quantitatively verifies the effectiveness of its feature extraction performance for face information processing. Simple-FLDA is an algorithm based on a geometrical analysis of the Fisher linear discriminant analysis. As a high-speed feature extraction method, the present algorithm in this paper is the improved version of Simple-FLDA. First of all, the approximated principal component analysis (learning by Simple-PCA) that uses the mean vector of each class is calculated. Next, in order to adjust within-class variance in each class to 0, the vectors in the class are removed. By this processing, it becomes high-speed feature extraction method than Simple-FLDA. The effectiveness is verified by computer simulations using face images.