Contents I. Introduction
SOMETIMES when we take pictures, we accidentally capture other people into the view of an image. For example, the girl in the image, who intended to take a solo picture, also captured another girl in the background. Similarly, the image of the couple in Fig. 1 included a random swinging person in the background. Initially these may be unnoticed but when you look at the photo afterwards, these distracting people in the background may have spoiled your visual experience. There has even been a term coined specifically for this called photobombing. One of the remedies is to crop these distracting people in the background out of the view of the image. However, hundreds of photos can be easily amassed over time and manually inspecting and cropping every photo will be tedious and time consuming. A good cropping also requires some amount of artistic skill to get a visually pleasing result. To address this problem, we present an automatic cropping system which automatically removes distracting people, emphasize the main subjects of interest, and improve the overall image composition. It uses heuristics such as background simplicity, rule of thirds combined with learned image representations from a convolutional neural network to produce an aesthetically pleasing image crop without the need of an artistic person.
