Sensation-based Photo Cropping

We train a quality classifier on photographs with user-provided ratings to assess how appealing a cropped region will be. We evaluate multiple cropping candidates and automatically select a region with a high quality score.

Cropping means cutting out part of a photograph. The choice of what to keep can change the impression it gives. This study proposes a method that automatically selects a crop that people are likely to find appealing.

We first train a scoring model using a large collection of photographs that people have rated on websites. For a new photograph, we create several possible crops, predict a quality score for each, and select the highest-scoring region. The key is to assess how the resulting photograph looks as a whole, rather than only checking whether it contains something eye-catching.

Existing methods focused on keeping regions that attract attention. However, keeping the most noticeable part does not always produce a pleasing photograph. Our method achieved better cropping results than the comparison method, which considered only the areas likely to attract the viewer's attention.

• Masashi Nishiyama, Takahiro Okabe, Yoichi Sato, and Imari Sato,
  Sensation-based Photo Cropping,
  Proceedings of ACM International Conference on Multimedia (ACM MM), October 2009.

• 西山 正志, 岡部 孝弘, 佐藤 洋一, 佐藤 いまり,
  複数の注目領域を用いた写真の主観的品質の識別,
  電子情報通信学会論文誌 D, Vol.J93-D, No. 8, pp. 1364 - 1374, 2010.

• Cropping result 1
• Cropping result 2