In this thesis, to construct an interface that displays similar images in the screen space by using some impression words, the possibility of classification of similar image group is described.
First the agreement of the similar image groups classified by subjective judgment of evaluator (subjective similar image group) and the groups classified with statistical method using impression degree (impression degree similar image group) was examined. Next to decide the user model the agreement of the impression degree between evaluators and the agreement of the images of the subjective similar image group were examined.
Twenty-one kinds of color images were chosen to classify the similar image groups and nine kinds of impression words were chosen.
An image classification system was developed to construct the similar image groups and to examine the impression degree of the images.
Thirty-five evaluators input their impression degree of each image with the system, and then those images were classified to the similar image groups. The results of obtaining are shown below.
(1) Because the agreement of the subjective similar image group and the impression degree similar image group was a range from 52 to 86%, a high agreement was proven.
(2) The main high correlation impression words among the evaluators were‘feminine’, ‘delicate’and ‘cheerful’and raw correlation words were‘simple’and‘retro’.
(3) The similarities of the evaluators were calculated by the principal co-ordinate analysis. As a result, 86% of the evaluators showed the same tendency.
Therefore, it was clarified that the classification of the similar image group using the impression words is possible, and the user model can cover a lot of users by constructing the representative model.
View full abstract