Hayato Itoh

Dimension Reduction and Construction of Feature Space for Image Pattern Recognition

We mathematically and experimentally evaluate the validity of dimension-reduction methods for the computation of similarity in image pattern recognition. Image pattern recognition identifies instances of particular objects and distinguishes differences among images. This recognition uses pattern recognition techniques for the classification and categorisation of images. In numerical 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. To ensure the accuracy of pattern recognition techniques, the dimension reduction of the vectors is an essential methodology since the time and space complexities of processing depend on the dimension of the data. Dimension reduction causes information loss of topological and geometrical features of image patterns. Through both theoretical and experimental comparisons, we clarify that dimension-reduction methodologies that preserve the topology and geometry in the image pattern space are essential for linear pattern recognition. For the practical application of methods of dimension reduction, the random projection works well compared with downsampling, the pyramid transform, the two-dimensional random projection, the two-dimensional discrete cosine transform and nonlinear multidimensional scaling if we have no a priori information on the input data.

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.

Global image registration by fast random projection

In this paper, we develop a fast global registration method using random projection to reduce the dimensionality of images. By generating many transformed images from the reference, the nearest neighbour based image registration detects the transformation which establishes the best matching from generated transformations. To reduce computational cost of the nearest nighbour search without significant loss of accuracy, we first use random projection. To reduce computational complexity of random projection, we second use spectrumspreading technique and circular convolution.

Multi-label classification for image annotation via sparse similarity voting

We present a supervised multi-label classification method for automatic image annotation. Our method estimates the annotation labels for a test image by accumulating similarities between the test image and labeled training images. The similarities are measured on the basis of sparse representation of the test image by the training images, which avoids similarity votes for irrelevant classes. Besides, our sparse representation-based multi-label classification can estimate a suitable combination of labels even if the combination is unlearned. Experimental results using the PASCAL dataset suggest effectiveness for image annotation compared to the existing SVM-based multi-labeling methods. Nonlinear mapping of the image representation using the kernel trick is also shown to enhance the annotation performance.

Approximation of N -Way Principal Component Analysis for Organ Data

We apply multilinear principal component analysis to dimension reduction and classification of human volumetric organ data, which are expressed as multiway array data. For the decomposition of multiway array data, tensor-based principal component analysis extracts multilinear structure of the data. We numerically clarify that low-pass filtering after the multidimensional discrete cosine transform efficiently approximates data dimension reduction procedure based on the tensor principal component analysis.

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.

Pattern recognition in multilinear space and its applications: mathematics, computational algorithms and numerical validations

We clarify the mathematical equivalence between low-dimensional singular value decomposition and low-order tensor principal component analysis for two- and three-dimensional images. Furthermore, we show that the two- and three-dimensional discrete cosine transforms are, respectively, acceptable approximations to two- and three-dimensional singular value decomposition and classical principal component analysis. Moreover, for the practical computation in two-dimensional singular value decomposition, we introduce the marginal eigenvector method, which was proposed for image compression. For three-dimensional singular value decomposition, we also show an iterative algorithm. To evaluate the performances of the marginal eigenvector method and two-dimensional discrete cosine transform for dimension reduction, we compute recognition rates for six datasets of two-dimensional image patterns. To evaluate the performances of the iterative algorithm and three-dimensional discrete cosine transform for dimension reduction, we compute recognition rates for datasets of gait patterns and human organs. For two- and three-dimensional images, the two- and three-dimensional discrete cosine transforms give almost the same recognition rates as the marginal eigenvector method and iterative algorithm, respectively.

Optical Flow Computation with Locally Quadratic Assumption

The purpose of this paper is twofold. First, we develop a quadratic tracker which computes a locally quadratic optical flow field by solving a model-fitting problem for each point in its local neighbourhood. This local method allows us to select a region of interest for the optical flow computation. Secondly, we propose a method to compute the transportation of a motion field in long-time image sequences using the Wasserstein distance for cyclic distributions. This measure evaluates the motion coherency in an image sequence and detects collapses of smoothness of the motion vector field in an image sequence.

Multiple Alignment of Spatiotemporal Deformable Objects for the Average-Organ Computation

We deal with multiple image warping, which computes deformation fields between an image and a collection of images, as an extension of variational image registration. Using multiple image warping, we develop a variational method for the computation of average images of biological organs in three-dimensional Euclidean space. The average shape of three-dimensional biological organs is an essential feature to discriminate abnormal organs from normal organs. There are two kinds of volumetric image sets in medical image analysis. The first one is a collection of static volumetric data of an organ and/or organs. The other is a collection of temporal volumetric data of an organ and/or organs. A collection of temporal volumetric beating hearts is an example of temporal volumetric data. For spatiotemporal volumetric data, we can compute (1) the temporal average, which is the average of a heart during a cycle, (2) the frame average, which is the average of hearts at a frame, and (3) the temporal average of frame averages.

Dimension reduction methods for image pattern recognition

In this paper, we experimentally evaluate the validity of dimension-reduction methods for the computation of the similarity in 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, 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 information used for classification.