In this paper, we propose a novel method for multi-class co-localization; it offers unsupervised localization of the main object in each image from an image set containing multiple kinds of main objects. Our method utilizes deep features to tackle the co-localization problem. Deep features, which can be extracted by pre-trained neural networks, are effective against unsupervised co-localization from multi-class image set. Based on spherical clustering, we classify deep features into several clusters, and choose one dominant cluster for each image or each image set. Experiments show that this very simple approach is significantly better than conventional state-of-the-art techniques in terms of localization accuracy. Moreover, multi-class co-localization experiments show that our method has the potential to classify the object in each image at the same time as achieving localization.
This paper introduces BumpMarker: a 3D-printed tangible marker that can perform simultaneous tagging, position tracking, and weight measurement of objects on pressure sensor sheets. The markers baseplate features several pins (raised dots) whose locations encode embedded information. A matrix pressure sensor sheet captures the pressure map of a marker-attached object on a sheet. The embedded data and object weight can be retrieved by processing the pressure map. We propose our design to achieve robust detection of the pins. We also show that our system has the ability to monitor weight changes in tagged objects. Through a series of evaluations, we investigate the technical feasibility of BumpMarker.
This paper aims to present an innovative design of motion estimation for sequential fisheye images. This design is an extended version of the original Lucas and Kanade's (LK) concept that used to design for calculating optical flow from general perspective images. The extended design consists of the LK concept and an additional self-improvement mechanism that automatically finds the maximum performance of the estimated motion. This extended scheme works much better than the original LK's idea or some block-based motion estimations. Moreover, to some extent, this proposed method is working extremely well to overcome some critical characteristics of the sequential fisheye images. These characteristics include distortion error on the fisheye image area, inconsistent brightness level, fluctuating number of object motion, changing the shape of object motion, or poor camera stability.
A novel method that integrates brain activity-based classifications obtained from multiple users is presented in this paper. The proposed method performs decision-level fusion (DLF) of the classifications using a kernelized version of extended supervised learning from multiple experts (KESLME), which is newly derived in this paper. In this approach, feature-level fusion of multiuser electroencephalogram (EEG) features is performed by multiset supervised locality preserving canonical correlation analysis (MSLPCCA). In the proposed method, the multiple classification results are obtained by classifiers separately constructed for the multiuser EEG features. Then DLF of these classification results becomes feasible based on KESLME, which can provide the final decision with consideration of the relationship between the MSLPCCA-based integrated EEG features and each classifier's performance. In this way, a new multi-classifier decision technique, which depends only on users' brain activities, is realized, and the performance in an image classification task becomes comparable to that of Inception-v3, one of the state-of-the-art deep convolutional neural networks.