An intelligent image cropping method and device based on explicit composition rule modeling

Abstract

The invention discloses an intelligent image cropping method and device based on explicit composition rule modeling, which belongs to the field of image aesthetics and multimedia technology. The method includes: i) using a deep convolution network as a skeleton network to extract features from an input image; ii ) For the composition branch, the confidence of each type of composition rule and the corresponding category activation map are generated based on the category activation mapping technology, and fused into a key element map; iii) For the clipping branch, the feature map of the skeleton network is further encoded, and the original The uniform anchor points are preset in the figure, and the clipping network outputs the offset between these anchor points and the marked cropping frame; each preset anchor point is weighted by using the key element map to obtain the predicted clipping frame; iv) use the loss function to The entire network is optimized. The invention pays attention to the composition interpretability of the clipping result, improves the composition quality of the clipping result, has high processing speed and strong real-time performance.

Description

technical field [0001] The invention belongs to the field of image aesthetics and multimedia technology, and more specifically relates to an intelligent image cropping method and device based on explicit composition rule modeling. Background technique [0002] With the improvement of living standards, people can no longer be satisfied with material pursuits, and more and more artistic and aesthetic elements have been integrated into people's lives. As an abstract cognitive concept, aesthetics is a huge challenge for computers to learn. With the development of digital photography and multimedia technology, more and more researches are focused on the fields related to aesthetics, aiming to make computers recognize aesthetics. [0003] Intelligent image cropping is one of the effective ways to improve the aesthetic effect of images, and it is widely used in the post-processing of photographic images. The purpose of image cropping is to cut off the redundant parts in the image...

AI Technical Summary

Problems solved by technology

However, none of the above methods can exploit the essential characteristics of the image cropping task, that is, the learning and encoding of composition rules and elements.

Images