This paper proposes construction technology for a Smart Tap Home Energy Management System (HEMS) that can make the energy consumption of living environment information and the consumer electronics visible using Wireless Sensor Networks (ZigBee). This technology solves the problem of the introduction cost of the HEMS spread Also, HEMS can be easily constructed in homes or the offices without modifying it to fit an existing building or consumer electronics product at all. Moreover, using the features of the ZigBee telecommunication, the technology is described that can display and analyze a detailed breakdown of the amount of electric power an hour. We developed and evaluated the proposed prototype Smart Tap HEMS for Indication. As a result, we found out how much electricity is wasted and shared the energy-saving strategy in our laboratory.

Energy consumption has been increasing over the past half century. Even though ultra-low power consuming appliances are being developed, the amount of energy consumption in homes is increasing because of various appliances being produced for ubiquitous smart home. We propose an adaptive home/building energy management system (A-HEMS/BEMS) for controlling energy consumption using the convergence of heterogeneous sensor/actuator networks. The system goes far beyond traditional auto meter reading (AMR) systems that offer only limited meter-reading functionality and provides a smart infrastructure management that brings new features and benefits to every aspect of our utilities.

The purpose of this study is to develop a flexible matching method for recognizing handwritten numerals based on the statistics of shapes and structures learned from learning samples. In the recognition method we reported before, there were problems in matching of the feature points and evaluation of matching. To solve them, we propose a new matching method supplementing contour orientations with convex/concave information and a new evaluation method considering the structure of strokes. With these improvements the recognition rate rose to 96.0% from the earlier figure 91.9%. We also made a recognition experiment on samples from the ETL-1 database and obtained the recognition rate 95.2%.

Corners in images include useful information and they play an important role in computer vision. However, stable and accurate comer detection is very difficult, because the edges composing a comer break easily around the corner. To overcome this difficulty, a corner detection method using slit rotational edge-feature detector (SRED) has been proposed. By evaluating the edge certainties at each pixel in all directions, this method can detect corners stably. Nevertheless, this method cannot detect the corners in the region of low contrast, and its detection accuracy depends on the edge directions. To solve these problems, we propose a new comer detection method using weighted and interpolated SIZED (WI-SRED). By using the edge certainties based on the region separability, the proposed method achieves stable corner detection in the region of low contrast, and moreover, by interpolating and weighting pixels, this method prevents the detection accuracy from depending on the edge directions.

The purpose of this study is to learn shapes and structures of a given learning set of handwritten numerals and to develop a flexible matching method for recognition based on the learning. First, this paper proposes a method of how to obtain a set of standard character patterns and the ranges of variations varying statistically from the given learning character samples. Then the recognition is made as follows: each standard pattern is deformed to match with the input character; and the matching is evaluated by the energy of deformation; and the closeness of the standard pattern to the input.

This paper proposes a robust method of detecting edge points by thresholding edge reliability with fixed threshold values. Traditionally edge detection is done by thresholding edge height. One crucial problem is that the threshold is determined ad hoc for the given input image. We use edge reliability instead of edge height, and show that two fixed thresholds work well. In our previous paper (1996) we proposed an edge vector extractor and stated the necessity of a measure, edge reliability, which can discriminate a low but steady edge from a noisy edge of large edge height. Our idea of extracting edge points is to detect points which are neither in a uniform region nor in a gradational region. The two detection thresholds, /spl radic/2//spl pi/ and /spl radic/3/4, are derived from the analysis of edge reliability in uniform regions and gradational regions contaminated by white Gaussian noises, respectively. Experiments show that our method can detect edge points stably, discriminating them from the points in uniform or gradational regions influenced by noises.

We propose a method of edge plane detection in 3D images by utilizing features of edge vector and edge reliability. This method can detect only reliable edge planes against noise influences. For the 2D case, we proposed an edge detection method which calculates edge vectors and edge reliability and can detect reliable edges at more accurate positions than Canny's method. This 2D edge vector detector is generalized to the 3D case. In applying this method to a 3D image, edge vectors and edge reliability are calculated. For finding edge surfaces (3D boundaries), the connected components of edge points detected from those features are composed by labeling. Finally, each edge surface is classified into edge planes using the directions of edge vectors. The proposed method was applied to spatio-temporal images and the depths of ridges in the 3D scene were estimated from the detected edge planes.

A scheme for learning and recognizing 3D objects from their 2D views is presented. The scheme proceeds in two stages. In the first stage, we try to learn a prototype automatically from 2D training images of different objects which belong to the same object class and consequently have similar shapes of parts and similar adjacency relations between the parts. In the second stage, the generated prototype is used to recognize learned objects or the objects similar to the learned ones from images of complex real scenes. We tested the approach on recognizing some simple objects from images of indoor scenes and got satisfactory results.

A scheme of generating a prototype automatically from multiple learning objects is presented. The learning objects are not just one and the same object but different objects belonging to a same object class. In this paper we show how to generate prototypes and test the generated prototypes in recognizing objects from images of real scenes. Satisfactory recognition results were obtained.

In this paper we propose a method to generalize a 1-D edge feature extractor to 2-D and clarifies its properties for the edges located closely each other (i.e., multiple edges). In a previous paper (1995) we stated necessity of a measure which can discriminate a clear edge with small edge height from a noisy edge with large edge height. Then we claimed that the edge features be viewed as a composition of edge height and edge reliability, based on the analysis of variances within a window around the edge. From this analysis, we constructed an edge feature extractor in 1-D. In this paper we generalize our 1-D edge feature extractor to a 2-D vectorized edge extractor and prove it able to calculate an edge orientation and edge height accurately and also to reduce computation time. Experiments show this clearly. The same vectorization technique is also applicable to the Canny's operator (1986). In multiple edges case, we made the conditions clear for calculating accurate edge locations. Since most edge detection methods use local maximums of the edge height function to detect edge points, we found the conditions by differentiating the edge height function.

A new approach is proposed to deal with the stereo matching problem. Based on the assumption that the scene is composed of some piecewise continuous surfaces, two stereo images can be considered to be composed of patches which can be matched continuously between two images. To find the matching of such patches, we divide the range of the depth in the scene into small intervals, and find continuous matchings of segments into the epipolar line in each small interval. Then the continuously matched segments are merged into patches, by considering the continuity between epipolar lines. The matchings of patches obtained in individual intervals, which may partially conflict or overlap with each other, are integrated to obtain the matchings of patches in the whole scene by considering the consistency among the matchings of patches.

An efficient graph matching approach is proposed for finding region correspondence between two images of the same scene but taken from different viewpoints. Regions and their relations in an image are represented with region adjacency graph (RAG), which is a kind of attributed planar graph. The problem to find an optimal region correspondence, which matches the regions in two images with maximal similarity in region features and region relations, is formulated into the problem to find the optimal inexact matching between two RAGs. The properties specific to planar graph and that of the region adjacency relations are utilized to invent an efficient algorithm to solve the problem. Experimental results on various kinds of images show the effectiveness of the method.>

A thinning algorithm is proposed for gray-scale images, which applies a combination of sequential and parallel conditions for pixel removal. The authors found that the algorithm developed by Salari and Siy (1984) produces unsuitable results in certain cases, due to its pure sequential processing. The proposed algorithm in this paper is an improvement of theirs and, at the same time, is a gray-image equivalent of Hilditch's thinning algorithm. Two versions are available for deriving either 4-connected or 1-connected core lines.>

An algorithm for the detection of dominant points and for building a hierarchical approximation of a digital curve is proposed. The algorithm does not require any parameter tuning and is shown to perform well for a wide variety of shapes, including scaled and rotated ones. Dominant points are first located by a coarse-to-fine detector scheme. They constitute the vertices of a polygon closely approximating the curve. Then, a criterion of perceptual significance is used to repeatedly remove suitable vertices until a stable polygonal configuration, the contour sketch, is reached. A highly compressed hierarchical description of the shape also becomes available.>

The authors report on the results of a comparison of seven methods for detecting corner points from given digital curves. In the experiments, they used two sets of input curves: flowchart symbol images and figures used in the past literature. Criteria for comparing the detection results are: extent of co-incidence with human results, processing time, and invariance to rotation, to size change, and to reflection. For flowchart symbol images, the method using n-codes produced the best result in the shortest processing time. However, for the figures from the literature Rosenfeld-Weszka's method (A. Rosenfeld, J.S. Weszka, 1975) worked best. No consistent conclusion has been made for invariance tests yet.>

A gray-scale thinning algorithm is proposed which is applicable to the case where a whole given image should be considered as the domain for thinning. The proposed algorithm is a modified version of the one proposed by the authors, which was intended for the case where a region to be thinned is specified in a given image. As an application of the method, extraction of a 'thin' edge curve from an edge-intensity image is presented.>

This paper proposes a new class of fusion operations which are useful for binary image processing. To begin with, an algorithm for the classical fusion operations is presented. They are sequentially composed of two basic operations, that is, expansion and contraction. The strategy of the proposed algorithm is as follows: the first operation is performed with the distance transformation, and the second operation utilizes the distance information obtained in the first operation. Next, a topological contraction and a topological expansion are defined. The topological contraction preserves the Euler number of an input image except that some simply connected components of 1 pixels are erased. The topological expansion preserves the Euler number of an input image except that some simply connected holes are filled. Then, several new fusion operations are obtained by combining the classical contraction and the topological expansion (or, the classical expansion and the topological contraction) sequentially. The algorithms for them use the same strategy as the proposed algorithm for the classical fusion operations. Finally, experimental results show that the new fusion operations and their combinations are effectively used in the binary image processing, such as shape smoothing, merging and separating connected components, and structure analysis.