View adjustment: helping users improve photographic composition
Nan Sheng,Yongzhen Ke,Shuai Yang,Yong Yang,Liming Chen
DOI: https://doi.org/10.1007/s00530-024-01490-x
IF: 3.9
2024-09-28
Multimedia Systems
Abstract:Image composition has always been one of the primary considerations for professional photography, and good composition can greatly enhance the perception of an image. However, it is difficult for most ordinary users to compose images quickly. To solve this challenge, we propose a new view adjustment method to improve image composition in this paper, assisting users in adjusting views promptly. We establish a composition feature extraction module, which can extract composition information of images based on professional photographic composition rules. Based on this module, we construct an image aesthetic assessment network to score the adjustment results. Additionally, we employ the concept of the hyper-net in our design to build a parameter generation-prediction network, enabling the application of view adjustment rules. In addition, we create a usable view adjustment dataset based on an existing image cropping dataset, which includes samples of displacement adjustment, scale adjustment, rotation adjustment, and their combined adjustment types to facilitate model training. Through experimentation, our model has demonstrated outstanding performance with an MAE score of 0.0392, an MSE score of 0.0037, and an IoU score of 0.783. Experimental results indicate that our proposed view adjustment method effectively improves the image composition, enabling real-time guidance for users to capture well-composed photographs.
computer science, information systems, theory & methods