Technological Development of Image Aesthetics Assessment

Abstract

Quantitative research on aesthetics is a classic interdisciplinary research. With the rapid development of deep learning, various approaches have been made in image aesthetics assessment (IAA). Starting from the concept of image aesthetics, this report roughly follows the chronological sequence and first introduces the manual design of image aesthetic features. We divide IAA into generic image aesthetics assessment (GIAA) and personalized image aesthetics assessment (PIAA) to introduce separately in the deep learning part. Majority of approaches are GIAA, which purpose is to simulate general aesthetics. In this section, we separately reviewed representative studies of five assessment methods (aesthetic classification, aesthetic regression, aesthetic distribution, IAA with attributes, aesthetic description). Due to the subjectivity of aesthetics, human’s aesthetics will more or less deviate from the generic value. PIAA aims to model the aesthetic preferences of specific user, and the research is of great value. We introduced this novel research in the fifth section. Finally, image aesthetic datasets of different uses are summarized. We hope this comprehensive survey can be helpful to researchers in the field of image and enhance the connection between computer and art.