Application of Image Blending in Multidimensional Persceptual Space -Taking Form of Automobiles as an Example

• Main Authors: Yang Shih-Hsun, 楊士勳
• Other Authors: Lin-Lin Chen
• Format: Others
• Language: zh-TW
• Published: 1998
• Online Access: http://ndltd.ncl.edu.tw/handle/13783018058098371471

碩士 === 國立臺灣科技大學 === 工程技術研究所 === 86 === This research applies image blending techniques to synthesize images that represent ideal points and cluster center points in a perceptual space, which is computed by using multidimensional scaling methods. The exterior styles of automobiles are used as stimuli in this research. This research started by extensively collecting side view images of automobiles, and common adjectives for describing the exterior styles of automobiles. Based on survey results, the collected automobiles and adjectives are clustered into main groups, from which 14 automobiles and 10 adjectives are selected as the stimuli and attributes, respectively, for the subsequent preference analysis. A preference survey, in which each subject is asked to rank each of the 14 automobiles according to 10 evaluative attr Next, the points representing the stimulus automobiles in the perceptual space are discretized by using Delaunay Triangulation, that maximizes the minimum angle. Any query point with the convex hull of the stimulus points thus lies within a unique triangle in this space. By connecting the query point to the corners of its corresponding triangle, three sub-triangles are obtained. The ratio of each of the smaller triangles to the original triangle can be used as the weights for the image blending. As such, aUsing the methods developed in this research, a full-colored image can be computed for ideal points and cluster center points in the perceptual space of automobiles. This tool can be used for studying marketing trends and for stimulating design ideas.