A Real Time Automatic Composition Algorithm for Mobile Terminal

Photo composition is one of the most basic skills in photography and image processing. Different composition methods often lead to different perceptual effects. Different application scenarios need different composition methods. For the scenes that need to extract main information such as website blog, good composition method is to select the most important information in the image, and for photography and other application scenes, good composition method is to keep the aesthetic feeling of the image and enhance the expressive power. In view of the need to keep the main body in composition and reduce the background area, this paper proposes an automatic composition algorithm based on lightweight salient object detection model and minimum information loss. The test is carried out on the CUHK composition data set and the Flick cropping composition data set. The IOU, model size and model running time are used as quantitative evaluation indexes, Compared with the mainstream aesthetic composition model, the experimental results show that the proposed method can achieve the same performance as the mainstream aesthetic based composition model, while the operation time is only one tenth of the aesthetic model, the model volume is only one percent of the aesthetic model, and the composition results are more stable, which is very suitable for the mobile terminal real-time automatic composition algorithm.