Kazuhiko Kawamoto

Saliency-based image editing for guiding visual attention

The most important part of an information system that assists human activities is a natural interface with human beings. Gaze information strongly reflects the human interest or their attention, and thus, a gaze-based interface is promising for future usage. In particular, if we can smoothly guide the users visual attention toward a target without interrupting their current visual attention, the usefulness of the gaze-based interface will be highly enhanced. To realize such an interface, this paper proposes a method for editing an image, when given a region in the image, to synthesize the image in which the region is most salient. Our method first computes a saliency map of a given image and then iteratively adjusts the intensity and color until the saliency inside the region becomes the highest for the entire image. Experimental results confirm that our image editing method naturally draws the human visual attention toward our specified region.

Learning dimensionality and orientations of 3D objects

Abstract We naturally classify plates as flat and boxes as of bulky structure. This understanding is based on the dimensionality of the objects. The dimensionality and orientation are important features for recognition of a 3D object, since these geometric properties are fundamental features for the classification of objects and for grasp-control for robots. In this paper, we derive a computational model for the classification of dimensionality of objects using properties of the mechanical moments of solid objects. Our model is based on the principal component analyzer (PCA) since the analyzer in the 3D Euclidean space derives directions of the mechanical moments of the objects from random samples. The directions of the principal components also determine the direction of objects. Therefore, our algorithm computes the orientations of objects in 3D space.

An efficient statistical method for subpixel optical flow detection

Abstract In this paper, we show that the randomized sampling and voting process detects optical flow. We introduce a random sampling method for solving the least-squares model-fitting problem using a mathematical property for the construction of pseudo-inverse. Using an appropriate number of images from a sequence of images, our method detects subpixel motion in this sequence. It is possible to compute subpixel motions from a long-time interval. We use the accumulator space for the unification of these flow vectors which are computed from different time intervals. Numerical examples for the test image sequences show the performance of our method.

Detection of spatial points and lines by random sampling and voting procedure

Abstract In this paper, we propose a method for the detection of spatial points and lines from a sequence of images. Our method does not require any predetermination of point correspondences among images. With camera motion, a sequence of images defines data in a spatiotemporal domain. In this domain, a trajectory of point correspondences among images defines a curve segment. For the detection of the curve segment in the spatiotemporal domain, we develop a classification process for points in the spatiotemporal domain where the camera motion is known. For the classification process, we adopt the voting procedure which is the main concept underlying the Hough transform.

Topology-Preserving Dimension-Reduction Methods for Image Pattern Recognition

In this paper, we experimentally evaluate the validity of dimension-reduction methods which preserve topology for image pattern recognition. Image pattern recognition uses pattern recognition techniques for the classification of image data. For the numerical achievement of image pattern recognition techniques, images are sampled using an array of pixels. This sampling procedure derives vectors in a higher-dimensional metric space from image patterns. For the accurate achievement of pattern recognition techniques, the dimension reduction of data vectors is an essential methodology, since the time and space complexities of data processing depend on the dimension of data. However, the dimension reduction causes information loss of geometrical and topological features of image patterns. The desired dimension-reduction method selects an appropriate low-dimensional subspace that preserves the topological information of the classification space.

Incorporating Audio Signals into Constructing a Visual Saliency Map

The saliency map has been proposed to identify regions that draw human visual attention. Differences of features from the surroundings are hierarchially computed for an image or an image sequence in multiple resolutions and they are fused in a fully bottom-up manner to obtain a saliency map. A video usually contains sounds, and not only visual stimuli but also auditory stimuli attract human attention. Nevertheless, most conventional methods discard auditory information and image information alone is used in computing a saliency map. This paper presents a method for constructing a visual saliency map by integrating image features with auditory features. We assume a single moving sound source in a video and introduce a sound source feature. Our method detects the sound source feature using the correlation between audio signals and sound source motion, and computes its importance in each frame in a video using an auditory saliency map. The importance is used to fuse the sound source feature with image features to construct a visual saliency map. Experiments using subjects demonstrate that a saliency map by our proposed method reflects human’s visual attention more accurately than that by a conventional method.

Shape Reconstruction from an Image Sequence

This paper clarifies a sufficient condition for the reconstruction of an object from its shadows. The objects considered are finite closed convex regions in three-dimensional Euclidean space. First we show a negative result that a series of shadows measured using a camera moving along a circle on a plane is insufficient for the full reconstruction of an object even if the object is convex. Then, we show a positive result that a series of pairs of shadows measured using a general stereo system with some geometrical assumptions is sufficient for full reconstruction of a convex object.

Global Image Registration Using Random Projection and Local Linear Method

The purpose of this paper is twofold. First, we introduce fast global image registration using random projection. By generating many transformed images as entries in a dictionary from a reference image, nearest-neighbour-search NNS-based image registration computes the transformation that establishes the best match among the generated transformations. For the reduction in the computational cost for NNS without a significant loss of accuracy, we use random projection. Furthermore, for the reduction in the computational complexity of random projection, we use the spectrum-spreading technique and circular convolution. Second, for the reduction in the space complexity of the dictionary, we introduce an interpolation technique into the dictionary using the linear subspace method and a local linear property of the pattern space.

Mathematical aspects of shape reconstruction from an image sequence

It is possible to decompose a three-dimensional objects into a collection of shadows. The geometric relation permits one to decompose shadows of a three-dimensional object to shadows of planar objects. Using this geometric relations, we prove that a class of non-convex objects is reconstructible from a series of shadows.

Random Sampling and Voting Method for Three-Dimensional Reconstruction

In the series of papers, we proposed a method for three-dimensional reconstruction from an image sequence without predetecting feature correspondences. In the method, we first collect all images and sample data, and second apply the reconstruction procedure. Therefore, the method is categorized into an off-line algorithm. In this paper, we deal with an on-line algorithm for three-dimensional reconstruction, if we sequentially measure images. Our method is based on the property that points and lines in space are uniquely computed from their projections between two images and among three images, respectively, if a camera system is calibrated. Using these property, our method determines both feature correspondences and three-dimensional positions of points and lines on an object.