What Are Your Impressions? Proposal for Emotional Assessment Platform

Abstract: Compelling visuals, spanning photographs, drawings, paintings, and illustrations, significantly enhance our daily encounters, conveying a wealth of information and eliciting emotions. Recognizing the impact of specific images and distinguishing them from less impactful ones proves valuable for various applications, including marketing and academic evaluations. However, evaluating these creative expressions presents inherent challenges. In our pursuit of understanding emotional responses to movie posters, we employed Russell's valence-arousal emotion model, incorporating measures of arousal and valence. Our investigation revealed that distinctive movie posters often received assessments marked by low arousal and positive valence. Expanding on these findings, we have enhanced the system to accommodate any image, incorporating administrative features tailored for experimental projects. This paper provides a concise overview of the system, emphasizing its capabilities, and explores a planned experiment use case.An overview of the system: Upon the successful login of a registered user into the system, the portal page is displayed, allowing the user to create a project. A project includes a title, description, and two measure titles, such as arousal and valence. Additionally, the project encompasses a collection of URLs corresponding to the images that need to be evaluated.Access to the evaluation pages is open to the public; therefore, there is no need to register an account to access them. The evaluation process consists of three steps: 1. Introduction, 2. Evaluation, and 3. Displaying the Evaluation Result. The introduction screen provides the project title and description to elucidate the purpose of image evaluation, accompanied by a 'Start Evaluation' button. Clicking this button directs users to the next page for the actual evaluation. On this page, six images, randomly selected from the database, are presented along with two evaluation scales ranging from -1.0 (negative value) to 1.0 (positive value). Users can assess their impressions of each image using two slider interfaces. Upon submitting the evaluation, the result page is generated, featuring a scatter plot graph that illustrates the user's decisions alongside the average scores calculated from other users for each image.Planned experiences: Conducting sensitivity evaluations for paintings involves examining people's ability to distinguish between human-created and AI-generated artworks, along with determining their preferences. The experiment utilized a set of images, comprising half human-drawn and half AI-generated, prompting participants to evaluate them based on two criteria: humanness and preference. The evaluators were unaware of the image authors. By plotting the results on two axes, we can analyze the preferred images by humans and AI and explore the discernibility of AI-created art.Similarly, a sensitivity evaluation for fashion is underway, utilizing two indices: fashionability and likability. Participants are asked to rate fashion items based on these criteria, shedding light on the styles deemed fashionable and preferred by individuals and providing insights into the variations in personal sensibilities. Additionally, collecting data on participants' age groups allows for an analysis of preferences across generations, thereby broadening the scope of the research.

Keywords: Emotional Evaluation, Emotional Responses on Images, Enhanced System for Image Assessment, Valence-Arousal Emotion Model

DOI: 10.54941/ahfe1005137

Cite this paper:

​