Toru Kurihara

Robust face detection using local CNN and SVM based on kernel combination

One key challenge of face detection is the large appearance variations due to some real-world factors, such as viewpoint, extreme illuminations and expression changes, which lead to the large intra-class variations and making the detection algorithm is not robust enough. In this paper, we propose a locality-sensitive support vector machine using kernel combination (LS-KC-SVM) algorithm to solve the above two problems. First, we employ the locality-sensitive SVM (LSSVM) to construct a local model on each local region, which can handle the classification task easier due to smaller within-class variation. Second, motivated by the idea that local features are more robust compared with global features, we use multiple local CNNs to jointly learn local facial features because of the powerful strength of CNN learning characteristic. In order to use this property of local features effectively, we apply the global and local kernels to the features and introduce the combination kernel to the LSSVM. Extensive experiments demonstrate the robustness and efficiency of our algorithm by comparing it with several popular face detection algorithms on the widely used CMU+MIT dataset and FDDB dataset.

Hierarchical prostate MRI segmentation via level set clustering with shape prior

Abstract Efficient and accurate segmentation of prostate is of great interest in image-guided prostate interventions and diagnosis of prostate cancer. In this paper, a novel hierarchical level set clustering approach is proposed to segment prostate from MR image, which makes full use of statistics information of manual segmentation result and incorporates shape prior into the segmentation task. The medium slice of prostate MR data, which is segmented artificially, is used to offer prior information and guide the segmentation of other slices. The Bhattacharyya coefficient between manual segmentation result of medium slice and local block region of pending slice is calculated to estimate the likelihood of local prostate region in pending slice. An adaptive blurring process is implemented before the optimization of level set function to restrain the redundancy texture information and retain the edge information in the meantime. We can capture the contour of prostate with a level set evolution embedded shape prior which is derived from the segmented result of medium slice. A comparative performance evaluation is carried out over a large set of experiments using real prostate magnetic resonance images and synthetic magnetic resonance data to demonstrate the validity of our method, showing significant improvements on both segmentation accuracy and noise sensitivity comparing to the state-of-the-art approaches.

Real-time 3D face modeling based on 3D face imaging

Abstract Traditional Iterative Closest Point (ICP) can not properly process the noise, outliers and missing data in face imaging, which would result in low accuracy of face image, face image registration error and much more noise in face image, to solve the above problems, an enhanced sparse ICP to register the 3D point clouds in face imaging is proposed. Sparse Iterative Closest Point (SICP) addressed these problems by formulating the registration optimization, which used sparsity inducing norms, moreover, a fast segmentation algorithm for head area segmentation in depth image was proposed. Based on the proposed fast segmentation algorithm and sparse ICP, a new real time 3D face modeling system was set up, which could generate real time 3D face models with high quality by using a depth camera (such as Kinect) even the background of face imaging was complicated.

Time division multiplexed orthogonal stripes pattern projection for single frame surface inspection system

We have proposed surface inspection system for detecting small defects on painted surface. The system used frequency multiplexed orthogonal stripes, so that we can detect anisotropic defects. But there is a little cross talk that affects detection rate of tiny defects. In this paper, we propose time division multiplexed orthogonal stripes pattern projection under single exposure. It is enabled by correlation image sensor.

Temporal modulated deflectometry for painted surface inspection

We present a fast method for measuring a curved specular surface defect, which is the temporal modulated deflectometry. The system uses correlation image sensor, which is developed by us. The correlation image sensor(CIS) outputs temporal correlation between intensity signal and reference signal. We moves rectangular pattern to generate temporal signal. There is no need to use sinusoidal intensity pattern for phase measuring deflectometry(PMD) because CIS captures only fundamental frequency component of rectangular wave projected on the screen. Hence, the methodology we proposed has a potential for fast inspection system using only single frame.

Real-time reflectance transformation using single frame surface orientation imager

In this paper, we propose real-time reflectance transformation system using correlation image sensor and four LEDs. Surface orientation of the object is encoded into amplitude and phase of the reflected light intensity by using phase shifted blinking LEDs, The correlation image sensor, provided by us, demodulates those amplitude and phase in each pixel during exposure time. Therefore, the surface orientation is captured by single frame, which can be applied moving object. We developed reflectance transformation system using surface orientation captured by our real-time surface orientation imager. We demonstrated that the system provides relighting and changing reflectance property in real-time.