加藤 直樹

We propose the prediction models which tell who will really purchase an apartment from among those who visit a model room, and analyze the characteristic of purchase based on data-mining technique. The data of questionnaire collected from those who visited model rooms usually has a lot of missing values. In order to improve the precision of prediction, the original data is converted to the one and two dimensional probability data concerning purchase. This data conversion technique is applied to the data which an estate agent in Kansai owns. Decision-tree and logistic regression are applied in order to obtain prediction models. Precisions of three different data types: original categorical data, one dimensional probability data and two-dimensional one are compared. As a result of experiment, the followings are observed: 1. two dimensional probability data with logistic regression has the highest precision through different data, 2. customers having high purchase probability do not answer the questionnaire too much.

This paper presents a floor layout algorithm for a two-story house based on enumeration algorithm as well as optimization algorithm. The process is as follows. First, all floor plans are created based on the floor enumeration algorithm and some of them which have higher values with respect to adjacent relationship and direction requirement are selected. Then, area and shape of those floor layouts are optimized by the tabu search. The proposed method has several advantages over existing methods; (1) Adjacency relationship among rooms is not fixed but the designer is allowed to explore all candidates based on enumeration algorithm of floor plans. (2) The enumeration algorithm employed here has found the all possible combination of 13 rooms. This number is enough to consider general floor layout of a house. (3) This method is applicable to a two-story house. A constraint of the two-story house is effectively used for searching solution in a short time.

A classifier for recognizing glass opening from architectural photographs based on extended Naive Bayes is proposed. From preliminary experiment we determine 3 possible candidates of attributes to represent color pixel each of which uses 6 color characteristics obtained together with x-y coordinates representing the relative position of the pixel in such a way that six color attributes are grouped into three pairs to be used in the extended Naive Bayes. Constructing the classifier based on the extended Naive Bayes which predicts whether a pixel represents glass or not, region segmentation is carried out based on MST. The computer experiments are carried out to demonstrate the effectiveness of the proposed method.

In this study, we propose a method for constructing an estimation model at the early stage of the building project which attains high accuracy and is robust. In the proposed method, the prediction accuracy is improved over the conventional method by focusing on the outliers and modifying the outliers's data when building the prediction model. More concretely, applying the so-called Winsorization mean and the trimmed mean which are known as statistical methods, the proposed methods appropriately modify the amount which is regarded as an outlier in constructing our prediction model. The method is then applied to two types of mutiple regression (L_2 regression and L_1 regression) and two types of neural network (multiple layer network and radial basis function network). In order to quantitatively evaluate the methods, they are applied to real building construction data. It is observed through the experiments that (i) the data modification is in general effective for improving the overall prediction accuracy, (ii) the Winsorization method is superior to the other data modification methods for any of four prediction models, and (iii) both of L_1 regression and the radial basis function network are superior to the other two.

This paper proposes two methods for identifying images which contain a particular architectural component from a database of architectural images. Applying the techniques of pattern recognition and image retrieval, the first method does this task by using the overall color distribution and the layout. Measuring the similarity between a target image and those whose classes are already known, the method determines the class which a target image belongs to, and it can be applied to classification for many types of architectural components. The second initially selects characteristic elements which will be supposed to play an important role in identifying a particular component, and constructs a model for recognizing such an architectural component by using those characteristic elements. Thus, this method requires us to construct a recognition model from scratch separately for each distinct architectural component, but it attains a higher recognition ability than the first one. The paper then carried out computational experiments by applying the methods to the identification of "tatami" and "lattice" (koshi in Japanese). It was observed that the proposed methods attains a certain level of accuracy and thus are useful for identifying architectural components.

Aiming at possible application to detection of several kinds of architectural objects such as windows from two-dimensional digital images, this paper proposes an efficient algorithm for detecting rectangular components in a binary edge image by limiting target objects to those having so-called slicing structure. The proposed algorithm first identifies the regions where ractangular objects may exist by projecting the image intensity data to two axes. It then further narrows the set of candidate edges that are supposed to form a part of rectangles by applying an efficient dynamic programming algorithm. It has been demonstrated through computational experiments that our algorithm attains much higher efficiency and detection accuracy compared with an existing method.

計算幾何学は,主として二次元平面や三次元空間における幾何学的データを対象としたデータ処理に関連する様々な問題を計算機を用いて高速かつ効率的に解くためのアルゴリズムを開発したり,その限界を究明する研究分野である.もともと,幾何学は古代建築などの基礎技術として発達してきたこともあり,建築と幾何学は古くから密接な関係がある.本稿では,空間骨組構造の設計問題から生じた一様三角形メッシュ生成問題について筆者らが取り組んだ研究成果の紹介を通して,多くの人に建築における計算幾何学の応用について理解を深め関心を持って頂くことをねらいとする.

This paper presents an optimization method for finding an optimal floor layout of rooms, passages and doorways in a possibly non-rectangular site, based on mathematical programming as well as genetic algorithm. The proposed method has several advantages over existing methods; (1) Adjacency relationship among rooms is not fixed but the designer is allowed to explore a number of candidates based on a sequence-pair and a genetic algorithm. (2) The shape of the building can be changed by introducing virtual rooms. (3) Optimization problem of the passage and doorway layouts is formulated as integer and linear program, respectively. We have implemented the algorithm and tested it for some example.

When one experinces an architectural space, he/she perceives various impressions. The purpose of this paper is to quantitatively clarify the relationship between the impression perceived on a photo of an architectural internal space and the phsical features of its color image. For fifty sample color images of internal space, we have performed a questionaire concerning what impression he/she acquires for each image by asking him/her to choose one of the impression words from a pair of antonyms. Also, we have computed color and texture features of photos. Here we used two-dimensional wavelet transform to obtain texture features while Lab-color space is used to extract color features. We then applied a decision-tree algorithm in order to derive interpretable and meaningful correlation of the impression words and image features. As a result, for images for which a majority of people had the same impression, we have found an interesting, interpretable common feature among the images.

Digital halftoning is a technique to convert a continuous-tone image into a binary image consisting of black and white dots. It is an important technique for printing machines and printers to output an image with few intensity levels or colors which looks similar to an input image. The purposes of this paper are to reveal that there are a number of problems related to combinatorial and computational geometry and to present some solutions or clues to those problems.

A method is presented for finding optimal prestresses of cable-supported frames as well as order of tensioning cables and removing temporary supports, which is refered to as construction order. The cable forces at the final state are first optimized, and the construction process is inversely traced. In the prestress optimization, prestresses of cables are modeled as external forces. Constraints are given for stresses and displacements of the frame. Two methods are presented for optimization of construction order of cable-supported frames considering constraints on stresses of the frame and cables. The globally optimal order is found by the dynamic programming approach. An approximate method is also presented based on a heuristic measure for selecting a cable or a temporary support at each step while tracing the inverse construction process only once.