| Summary: | 碩士 === 國立臺灣科技大學 === 設計研究所 === 91 === Designers often use different kinds of image data. Now the most convenient way of collecting them is mainly from the Internet or Image CD. However, with the rapid increase of image material, the users are faced with the problem that the amount of data is so huge that makes searching difficult. Therefore, it has been became an urgent need to promote the efficiency of query
.
This study puts forward a method that used “Feelings" to query image data as well as aids the current image query so as to attain the effect of complementing each other. The analysis of image information could be divided into two different levels: one level means the described affairs or objects; the other deals with people’s feelings towards these affairs or objects. However, the current query system puts great emphasis on the former, but pays little attention to the latter. Therefore, this study discusses the possibility of querying image by “Feelings” as the following three phases.
Firstly, aiming at various kinds of image CD products from market, we collect the feelings from them. Then through Cluster Analysis, the collection summed up as 12 representative phrases of feelings that are widely used and could well represent general image data, and all these 12 phrase sets would be used as query of the emulation database.
Secondly, we obtain the subject weights of emulation database by the experiment of coupling images and feelings, and we discuss the relationship between the image characters and feelings. The result shows that the distribution of these subject weights was in accord with general databases. In the field of feelings, different feelings primarily depended on the theme or tone of the images.
Finally, the study assesses the efficiency of this kind of query and discusses whether cognition of system management and searchers has something in common. The key points of the assessment are Prescision Ratio and Recall Ratio, and the result was 0.5366 and 0.406 respectively, beyond the weight level of 0.6. Through the examination of Bivariate Correlations between the subject weights and the times of query of images, the result shows that the direct correlation obviously exists and this kind of method really contains its feasibility.
|
|---|
