@inproceedings{Wit2022,
Author = {Withöft, Ani and Abdenebaoui, Larbi and Boll, Susanne},
Title = {ILMICA - Interactive Learning Model of Image Collage Assessment: A Transfer Learning Approach for Aesthetic Principles},

Year = {2022},
Pages = {84-96},
Editor = {Springer International Publishing},
Publisher = {Springer Nature Switzerland AG},
Series = {Lecture Notes in Computer Science, Volume 13142},
Isbn = {978-3-030-98354-3},
Booktitle = {International Conference on Multimedia Modeling},
Doi = {https://doi.org/10.1007/978-3-030-98355-0_8},
Url = {https://link.springer.com/10.1007/978-3-030-98355-0_8},
type = {inproceedings},
Abstract = {The beauty of moments can be expressed in many ways. One of them is the image collage which captures events and expresses emotions. Nowadays there is a large number of digital images. Aesthetic analyses of image collages are rarely performed due to their complexity and time-consuming nature. For this reason, this is an important issue that has to be addressed. In this paper, we propose an interactive learning model for image collage assessment. It consists of two components: A pre-trained convolutional neural network with built-in knowledge about aesthetics obtained from single image analysis, and an “Interactive Transfer Learning” component specialized in collage aesthetics which is adaptable via Active Learning. We present a mixed method study in which rules for software-based collage generation are identified and a dataset of automatically generated collages representative of the rules is created. ILMICA’s performance is analyzed by a user survey. It is found that the knowledge transfer from single image assessment to collage assessment works: ILMICA can assess collage aesthetics based on predefined rules, thereby demonstrating the system’s ability to learn. Thus, this process can alleviate the end user and simplify aesthetic collage evaluations.}
}
@COMMENT{Bibtex file generated on }