Vol.1 No.1 2008

47/76

Research paper : Technologies for the design and retail service of well-fitting eyeglass frames (M. Mochimaru, et al.)−44−Synthesiology - English edition Vol.1 No.1 (2008) the validation study were equivalent to those of the main experiment. The residual error between estimated ratings and actual ratings for validation images were calculated for images of (1) and (2). The error was within 10 % for all impression ratings. Moreover, a similar error range was observed in the deformed actual face images of (3). It was confirmed that the computational model could be utilized for real face images. The hypothesis of this study, that impression rating can be estimated by a function of shape factors of the face and eyeglass frame, was thus validated.3.5 Style recommendation systemA style recommendation system based on the computational model to estimate impression ratings (Section 3.4) was developed. This is part of the recommendation system in Figure 2, but focuses on style recommendation using 2-D face images. Customer images are captured by a USB camera connected to a PC, and a 2-D homologous face model (Figure 8) is generated automatically. In the present system, the interpupillary distance is measured directly and the value is used for scaling. Scaling problems can also be solved by camera calibration in the future. Images of 12 eyeglass frames used in the experiment and their shape factors are registered in the system. Obtaining a 2-D model of the customer, the impression ratings of all pairs of adjectives for 12 eyeglass frames are calculated immediately. The ratings are presented on the system as a relative scale for the 12 frames with a composite image of the face and selected frame (Figure 10).4 Integration of elemental technologiesThe aim of this study is to realize a solution for “Finding Well-fitting Products” through an application study aiming to fit eyeglass frames to individual customers. Elemental technologies were developed for this purpose and they were integrated to realize the solution presented in Figure 2. Three elemental technologies described in Section 3, size variation design, head shape reconstruction, and estimation of impression ratings, were integrated into an off-line system. Part of this system was completed as a demonstration.The most important foundation in this study is the database of 3-D face and head shapes. Homologous modeling and the database are essential for integration. All elemental technologies developed in this study require the database. For instance, the head shape reconstruction method requires the database obtained using another measurement method. The reconstruction method itself is not complete without the database of homologous models of head shape. This promotes the advantage for system providers who have a database. The system provider can keep the advantage of the database by applying the system. When the provider applies the system at a retail shop, the database can continuously be expanded by adding customer data to the database. The compiled data can be used for both proper size grouping of products, as well as for robust shape reconstruction using a multi-camera system.5 DiscussionIn this section, the industrial and social validity of this study are discussed. Based on the size variation method described in Section 3.2, new eyeglass frames with proper sizing were developed by the collaborating company and were sold on the market. Subsequently, the company used the sizing method for other eyeglass frames. It was thus confirmed that the method is valid for industry.The 3-D shape reconstruction method described in Section 3.3 has a similar function as 3-D shape scanning technologies, but with this method, complete 3-D shapes can be obtained very quickly and without laser scanning. The error of 2 mm is larger than the 0.5 mm of conventional scanning technologies [3], but is small enough for size and style recommendation. The system contains multiple digital cameras, and registration and time control for projectors and cameras are not required. The features of the system, such as its low cost, size, lack of laser projection, and quick and complete measurement, are superior to conventional systems for retail use. The style recommendation system described in Sections 3.4 and 3.5 is only used for demonstration purposes. This system has been received well by visitors, with comments such as “it is fun to select eyeglass frames with this system”.Throughout this study, the obstacles to investment from providers were reduced by the development of efficient size variation designs using existing manufacturing resources, Fig. 10 Style recommendation system for eyeglass frames based on Kansei (sensibility) modeling.

※このページを正しく表示するにはFlashPlayer9以上が必要です