2020/09/08 by Prathmesh Madhu, Madhu, Prathmesh, Tilman Marquart +9 · 3 citations
Neuroscience · Computer Science · Earth and Planetary Sciences · #Aesthetic Perception and Analysis #Visual Attention and Saliency Detection #3D Surveying and Cultural Heritage
paper · pdf · doi:10.48550/arxiv.2009.03807
Image compositions as a tool for analysis of artworks is of extreme\nsignificance for art historians. These compositions are useful in analyzing the\ninteractions in an image to study artists and their artworks. Max Imdahl in his\nwork called Ikonik, along with other prominent art historians of the 20th\ncentury, underlined the aesthetic and semantic importance of the structural\ncomposition of an image. Understanding underlying compositional structures\nwithin images is challenging and a time consuming task. Generating these\nstructures automatically using computer vision techniques (1) can help art\nhistorians towards their sophisticated analysis by saving lot of time;\nproviding an overview and access to huge image repositories and (2) also\nprovide an important step towards an understanding of man made imagery by\nmachines. In this work, we attempt to automate this process using the existing\nstate of the art machine learning techniques, without involving any form of\ntraining. Our approach, inspired by Max Imdahl's pioneering work, focuses on\ntwo central themes of image composition: (a) detection of action regions and\naction lines of the artwork; and (b) pose-based segmentation of foreground and\nbackground. Currently, our approach works for artworks comprising of\nprotagonists (persons) in an image. In order to validate our approach\nqualitatively and quantitatively, we conduct a user study involving experts and\nnon-experts. The outcome of the study highly correlates with our approach and\nalso demonstrates its domain-agnostic capability. We have open-sourced the code\nat https://github.com/image-compostion-canvas-group/image-compostion-canvas.\n