Abstract :
Emotional meaning is critical for users to retrieve relevant images. However, because emotional meanings are subject to the individual viewerʹs interpretation, they are considered difficult to implement when designing image retrieval systems. With the intent of making an imageʹs emotional messages more readily accessible, this study aims to test a new approach designed to enhance the accessibility of emotional meanings during the image search process. This approach utilizes image searchersʹ emotional reactions, which are quantitatively measured. Broadly used quantitative measurements for emotional reactions, Semantic Differential (SD) and Self-Assessment Manikin (SAM), were selected as tools for gathering usersʹ reactions. Emotional representations obtained from these two tools were compared with three image perception tasks: searching, describing, and sorting. A survey questionnaire with a set of 12 images was administered to 58 participants, which were tagged with basic emotions. Results demonstrated that the SAM represents basic emotions on 2-dimensional plots (pleasure and arousal dimensions), and this representation consistently corresponded to the three image perception tasks. This study provided experimental evidence that quantitative usersʹ reactions can be a useful complementary element of current image retrieval/indexing systems. Integrating usersʹ reactions obtained from the SAM into image browsing systems would reduce the efforts of human indexers as well as improve the effectiveness of image retrieval systems.
