An image intelligent cropping method and system based on confrontational domain adaptation

Abstract

The invention provides an image intelligent cropping method and system based on confrontational domain adaptation, which belongs to the field of computer vision. The method specifically includes: inputting a target domain image to be cropped in a target application scene to a trained feature extractor to obtain global features ;According to the preset clipping method, the global features are resampled; the regional features are input to the aesthetic classifier for aesthetic scoring, and the clipping results are screened; the training process of the feature extractor is: the domain adaptation loss gradient calculated based on the target domain samples is reversed After transfer, it is transmitted to the feature extractor, and the domain adaptation loss gradient calculated based on the source domain samples is kept unchanged and transmitted to the feature extractor. The feature extractor learns the ability to align the global features of the source domain and target domain samples; and adjusts itself according to the aesthetic loss. Parameters, learn the ability of aesthetic analysis; the training of aesthetic classifier is to adjust its own parameters according to the aesthetic loss. The invention solves the problem that the performance of the existing intelligent clipping method drops significantly during the cross-domain test.

Description

technical field [0001] The invention belongs to the field of computer vision, and more specifically relates to an image intelligent cropping method and system based on adversarial domain adaptation. Background technique [0002] The task of intelligent cropping is to crop out a region with a more reasonable composition from the original image as a new generated image, and replace the low-quality original image with a high-quality generated image. Because it only involves cropping and is easy to operate, it can significantly improve the aesthetic quality of images. Image cropping is widely used as an image editing operation. However, manual cutting requires a certain technical threshold, which is not only time-consuming, but also a skill that not everyone can master. Therefore, it makes sense to use intelligent algorithms to automatically crop. Compared with manual cropping, smart cropping does not require users to have a high aesthetic quality, nor does it need to understa...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the defects of the prior art, the purpose of the present invention is to provide an image intelligent cropping method and system based on confrontational domain adaptation, which aims to solve the problem of cross-domain and cross-data The problem that the performance of the clipping model drops significantly when the set is tested

Images