Shinichiro Omachi

Structural, Syntactic, and Statistical Pattern Recognition

Peer-to-Peer (P2P) lending is an online platform to facilitate borrowing and investment transactions. A central problem for these P2P platforms is how to identify the most influential factors that are closely related to the credit risks. This problem is inherently complex due to the various forms of risks and the numerous influencing factors involved. Moreover, raw data of P2P lending are often high-dimension, highly correlated and unstable, making the problem more untractable by traditional statistical and machine learning approaches. To address these problems, we develop a novel filter-based feature selection method for P2P lending analysis. Unlike most traditional feature selection methods that use vectorial features, the proposed method is based on graphbased features and thus incorporates the relationships between pairwise feature samples into the feature selection process. Since the graph-based features are by nature completed weighted graphs, we use the steady state random walk to encapsulate the main characteristics of the graphbased features. Specifically, we compute a probability distribution of the walk visiting the vertices. Furthermore, we measure the discriminant power of each graph-based feature with respect to the target feature, through the Jensen-Shannon divergence measure between the probability distributions from the random walks. We select an optimal subset of features based on the most relevant graph-based features, through the Jensen-Shannon divergence measure. Unlike most existing state-of-theart feature selection methods, the proposed method can accommodate both continuous and discrete target features. Experiments demonstrate the effectiveness and usefulness of the proposed feature selection algorithm on the problem of P2P lending platforms in China.

A handwritten character recognition system using directional element feature and asymmetric Mahalanobis distance

This paper presents a precise system for handwritten Chinese and Japanese character recognition. Before extracting directional element feature (DEF) from each character image, transformation based on partial inclination detection (TPID) is used to reduce undesired effects of degraded images. In the recognition process, city block distance with deviation (CBDD) and asymmetric Mahalanobis distance (AMD) are proposed for rough classification and fine classification. With this recognition system, the experimental result of the database ETL9B reaches to 99.42%.

Traffic light detection with color and edge information

The concern of the intelligent transportation system rises and many driver support systems have been developed. In this paper, a fast method of detecting a traffic light in a scene image is proposed. By converting the color space from RGB to normalized RGB, some regions are selected as candidates of a traffic light. Then a method based on the Hough transform is applied to detect an exact region. Experimental results using images including a traffic light verifies the effectiveness of the proposed method.

Fast Template Matching With Polynomials

Template matching is widely used for many applications in image and signal processing. This paper proposes a novel template matching algorithm, called algebraic template matching. Given a template and an input image, algebraic template matching efficiently calculates similarities between the template and the partial images of the input image, for various widths and heights. The partial image most similar to the template image is detected from the input image for any location, width, and height. In the proposed algorithm, a polynomial that approximates the template image is used to match the input image instead of the template image. The proposed algorithm is effective especially when the width and height of the template image differ from the partial image to be matched. An algorithm using the Legendre polynomial is proposed for efficient approximation of the template image. This algorithm not only reduces computational costs, but also improves the quality of the approximated image. It is shown theoretically and experimentally that the computational cost of the proposed algorithm is much smaller than the existing methods.

Detection of traffic light using structural information

Driver support systems using images are paid attention and various researches on recognizing and understanding the road environment have been done. If it is possible to detect and recognize traffic lights, it will give useful information to a driver to understand the road environment. In this paper, a method of detecting a traffic light in a scene image is proposed. Considering the structure of a traffic light, we propose a method for detecting a traffic light based on the Hough transform. Experimental results using images including a traffic light taken by a digital camera through a windshield verifies the effectiveness of the proposed method.

A fast algorithm for a k‐NN classifier based on the branch and bound method and computational quantity estimation

The nearest neighbor rule or k-nearest neighbor rule is a technique of nonparametric pattern recognition. Its algorithm is simple and the error is smaller than twice the Bayes error if there are enough training samples. However, it requires an enormous amount of computation, proportional to the number of samples and the number of dimensions of the feature vector. In this paper, a fast algorithm for the k-nearest neighbor rule based on the branch and bound method is proposed. Moreover, a new training algorithm for constructing a search tree that can reduce the computational quantity is proposed. Experimental results show the effectiveness of the proposed algorithms.

A noise-adaptive discriminant function and its application to blurred machine-printed Kanji recognition

Accurate recognition of blurred images is a practical but previously mostly overlooked problem. In the paper, we quantify the level of noise in blurred images and propose a modification of discriminant functions that adapts to the level of noise. Experimental results indicate that the proposed method actually enhances the existing statistical methods and has impressive ability to recognize blurred image patterns.

Tooth shape reconstruction from ct images using spline Curves

It is desired to obtain three-dimensional shapes of teeth without extracting them physically in order to construct teeth database. In this paper, we propose a method for reconstructing three-dimensional shape of a tooth from the images acquired by a dental micro CT. First, initial contour of a tooth is obtained from an image in which the tooth is not buried in the bone and does not touch another tooth. A contour is determined in the adjacent image by searching the positions where the edge is strong. The shape of the tooth is obtained by recursive detection of the contours. Active contour model is used to obtain the contour where the edge is strong, and closed spline curve is used to represent the contour. Experimental results show that the shape of a tooth can be reconstructed from actual CT images by the proposed method.

Structure extraction from decorated characters using multiscale images

Decorated characters are widely used in various documents. Practical optical character reader is required to deal with not only common fonts but also complex designed fonts. However, since the appearances of decorated characters are complicated, most general character recognition systems cannot give good performances on decorated characters. In this paper, an algorithm that can extract characters essential structure from a decorated character is proposed. This algorithm is applied in preprocessing of character recognition. The proposed algorithm consists of three procedures: global structure extraction, interpolation of structure and smoothing. By using multiscale images, topographical features, such as ridges and ravines are detected for structure extraction. Ridges are used for extracting global structure and ravines are used for interpolation. Experimental results show character structures can be clearly extracted from very complex decorated characters.

A New Approximation Method of the Quadratic Discriminant Function

For many statistical pattern recognition methods, distributions of sample vectors are assumed to be normal, and the quadratic discriminant function derived from the probability density function of multivariate normal distribution is used for classification. However, the computational cost is O(n2) for n-dimensional vectors. Moreover, if there are not enough training sample patterns, covariance matrix can not be estimated accurately. In the case that the dimensionality is large, these disadvantages markedly reduce classification performance. In order to avoid these problems, in this paper, a new approximation method of the quadratic discriminant function is proposed. This approximation is done by replacing the values of small eigenvalues by a constant which is estimated by the maximum likelihood estimation. This approximation not only reduces the computational cost but also improves the classification accuracy.