Takio Kurita
Hiroshima University, Information Engineering, Faculty Member
- I joined the Electrotechnical Laboratory, AIST, MITI in 1981. From 1990 to 1991, I was a visiting research scientist ... moreI joined the Electrotechnical Laboratory, AIST, MITI in 1981. From 1990 to 1991, I was a visiting research scientist at Institute for Information Technology, National Research Council Canada. From 2001 to 2009, I was a deputy director of Nueroscience Research Institute, National Institute of Advanced Industrial Science and Technology (AIST). Also, I was a Professor at the Graduate School of Systems and Information Engineering, University of Tsukuba, from 2002 to 2009. I am currently a Professor at Hiroshima University. My current research interests include statistical pattern recognition and its applications to image recognition.edit
Research Interests:
For dimensionality reduction, t-distributed stochastic neighbor embedding (t-SNE) is famous. This technique represents the similarity between the pair of the samples in the high-dimensional space as Gaussian distribution. Then the... more
For dimensionality reduction, t-distributed stochastic neighbor embedding (t-SNE) is famous. This technique represents the similarity between the pair of the samples in the high-dimensional space as Gaussian distribution. Then the similarity between the pair of the samples in the low-dimensional embedding space is also represented by using t-distribution to obtain the feature vectors in the embedding space from the high-dimensional data. The authors proposed q-Gaussian distributed stochastic neighbor embedding (q-SNE) as an extension of the t-SNE. The q-Gaussian distribution can express many distributions by setting hyperparameter q$= and includes the Gaussian distribution and the t-distribution as the special cases with hyperparameter q close to 1.0 and q = 2.0. However, these methods are applicable for a given data set and it is not possible to map new samples into the embedded space. To address this problem, the parametric t-SNE is proposed to construct the non-linear mapping by using a feed-forward neural network. In this paper, we propose a novel technique called parametric q-SNE with Convolutional Neural Network without pre-training. On MNIST, FashionMNIST, and COIL-20, the effectiveness of the parametric q -SNE is shown by using the visualization on 2-dimensional mapping, and the classification by using k nearest neighbors (k-NN) on the embedded space.
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Abstract In this study, we propose a novel autoencoder framework based on orthogonal projection constraint (OPC) for anomaly detection (AD) on both complex image and vector datasets. Orthogonal projection is useful to capture the null... more
Abstract In this study, we propose a novel autoencoder framework based on orthogonal projection constraint (OPC) for anomaly detection (AD) on both complex image and vector datasets. Orthogonal projection is useful to capture the null subspace that consists of noisy information for AD, which is explicitly ignored in the existing approaches. The exploration of double subspaces, called normal space (NS) and abnormal space (AS) can improve the discriminative manifold information. Therefore, in this study, autoencoder framework based on the OPC learning method is proposed that combines the orthogonal subspace score and the reconstruction error score in the target tasks for AD. To the best of our knowledge, this is the first study that introduces an autoencoder-based model with two orthogonal subspaces for AD. Through the orthogonality, the anomaly-free data and abnormal ⧹ nosiy information are projected into the NS and the AS, respectively. Thus, it potentially addresses the problem of the distribution of generative model by combining the abilities of two subspaces that can appropriately learn the features and establish a strict boundaries around the normal data. For image datasets, we propose a convolutional autoencoder based on OPC. Additionally, the generalization and adaptability of the proposed method in AD was investigated using vector datasets by implementing a fully-connected layer-based OPC in the encoder-decoder structure. The effectiveness of the proposed framework for AD was evaluated through the comparison with state-of-the-art approaches.
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
This paper describes a method for calibrating non-overlapping cameras in a simple way: using markers on the cameras. By adding an AR (Augmented Reality) marker to a camera, we can find the transformation between the fixed AR marker and... more
This paper describes a method for calibrating non-overlapping cameras in a simple way: using markers on the cameras. By adding an AR (Augmented Reality) marker to a camera, we can find the transformation between the fixed AR marker and the camera's center. With such information, relative pose of cameras can be easily found as long as the marker located on them is visible. Our method consists of the following two steps: (1) use an extra camera and a chessboard to find the transformation between the AR marker and the camera center. (2) Use the information from (1) and transformation between different markers to calibrate non-overlapping cameras. Compare to other non-overlapping calibration methods, our method can work with as less as one image and does not suffer from the degenerate cases.
Research Interests:
In this paper we propose a method able to automatically detect good/bad colonies of iPS cells using local patches based on densely extracted SIFT features. Different options for local patch classification based on a kernelized novelty... more
In this paper we propose a method able to automatically detect good/bad colonies of iPS cells using local patches based on densely extracted SIFT features. Different options for local patch classification based on a kernelized novelty detector, a 2-class SVM and a local Bag-of-Features approach are considered. Experimental results on 33 images of iPS cell colonies have shown that excellent accuracy can be achieved by the proposed approach.
Research Interests:
Research Interests:
Research Interests:
Research Interests:
In this paper we extend the autoregressive (AR) model to the multilevel AR model with wavelet transformation, in order to get the AR coefficients at each level as a set of shape descriptors for every level. To get the multilevel AR model,... more
In this paper we extend the autoregressive (AR) model to the multilevel AR model with wavelet transformation, in order to get the AR coefficients at each level as a set of shape descriptors for every level. To get the multilevel AR model, we use the wavelet transformation such as Haar wavelet to a boundary data. Then real AR and complex-AR (CAR) models are adopted to the multilevel boundary data of a shape to extract the features at each level. Furthermore we present the relation of the autocorrelation coefficients between adjacent resolution levels to elucidate the relation between AR model and wavelet transformation. Some experiments are also shown for the multilevel AR and CAR models with a certain similarity measure.
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Segmentation of teeth in Cone-Beam Computed Tomography (CBCT) images is challenging problem due to its noise and the similar grayscale intensity of bone and teeth element. In this paper we proposed a new method based on three-dimensional... more
Segmentation of teeth in Cone-Beam Computed Tomography (CBCT) images is challenging problem due to its noise and the similar grayscale intensity of bone and teeth element. In this paper we proposed a new method based on three-dimensional (3D) region merging and histogram thresholding for automatic segmentation of teeth on CBCT images. The proposed 3D region merging algorithm can recognized the teeth element that have similar intensity with the bone element based on the three-dimensional (3D) information of the neighboring slices of the CBCT image. Merging the teeth region will lead to more homogenous grayscale intensity distribution inside the teeth. Then histogram thresholding that utilized the characteristic of CBCT images is performed to binarize the grayscale images and obtain the teeth object. The average accuracy, sensitivity, and specificity of the proposed method are 97.75%, 80.22%, and 98.31%, respectively. The proposed method is fully automatic, therefore lead to more objective and reproducible results.
Research Interests:
Video Super resolution algorithms usually utilize the motion information of each pixel in consecutive frames to interpolate pixel values in a higher resolution and reconstruct frames of a higher resolution video from a lower resolution... more
Video Super resolution algorithms usually utilize the motion information of each pixel in consecutive frames to interpolate pixel values in a higher resolution and reconstruct frames of a higher resolution video from a lower resolution input video. Recently, architectures based on deep neural networks have gained popularity and can generate higher resolution videos with better visual quality. Also, deep neural networks make single image super resolution possible. In single image super resolution, a higher resolution image is constructed from a given input image by learning the transformation from lower resolution images to the higher resolution images. In this paper we propose to apply the single image super resolution algorithm to each frame in the original video and then use the resulting frames to estimate the movements of each pixel. Since the spatial resolution of the estimated motion of each pixel affects the visual quality of the higher resolution video, it is expected that the proposed approach can improve the visual quality of video super resolution. The proposed approach consistently results in good quality video reconstruction when tested on videos with diverse contents and different motion levels, which outperforms state of the art algorithms and offers competing visual performance on benchmark datasets.
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests: Biology and Regression
Research Interests:
Research Interests:
This paper introduces the sparse regularization for the convolutional neural network (CNN) with the rectified linear units (ReLU) in the hidden layers. By introducing the sparseness for the inputs of the ReLU, there is effect to push the... more
This paper introduces the sparse regularization for the convolutional neural network (CNN) with the rectified linear units (ReLU) in the hidden layers. By introducing the sparseness for the inputs of the ReLU, there is effect to push the inputs of the ReLU to zero in the learning process. Thus it is expected that the unnecessary increase of the outputs of the ReLU can be prevented. This is the similar effect with the Batch Normalization. Also the unnecessary negative values of the inputs of the ReLU can be reduced by introducing the sparseness. This can improve the generalization of the trained network. The relations between the proposed approach and the Batch Normalization or the modifications of the activation function such as Exponential Linear Unit (ELU) are also discussed. The effectiveness of the proposed method was confirmed through the detail experiments.