Data enhancement method and system for composite image

Abstract

The invention relates to the technical field of image data processing, in particular to a data enhancement method and system for a composite image, and the method comprises the following steps: an image collection step: collecting an image through an image collection module; an encoder training step: training four groups of different encoders according to the collected images; a model generation step: coding combinations in the four groups of encoders are input into the GAN model, and then a synthesized image is generated; an image comparison step: comparing the synthesized image with an original image, and then calculating four groups of loss values of the synthesized image; and a judgment generation step: inputting the synthesized image into a GAN model for judgment. According to the method, the image is decomposed into four elements including the scene, the shape, the posture and the texture to be coded and represented, and then different generated images are formed through different combinations of the four elements, so that the diversity of the image or the target object is greatly improved, and the accuracy of a model generation result is improved.

Description

technical field [0001] The invention relates to the technical field of image data processing, in particular to a data enhancement method and system for composite images. Background technique [0002] In the process of image acquisition, due to the limited environment or cost at the time of acquisition, the scene where the target is actually collected, the shape, posture and texture (or color) of the target are very limited, and currently affect the actual effect of deep learning. Among the three major factors of computing power, data, and models, data constraints are often the most common. The diversity of target images caused by such data is limited, which will make the final effect or generalization ability of model training poor. Common solutions include enhancing data, such as changing geometry, color, or combining multiple images to generate corresponding training data to enhance the generalization ability of the model. [0003] Chinese Patent No. CN112132133A provide...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to provide a data enhancement method and system for composite images to solve the problems that the existing image data enhancement methods proposed in the above-mentioned background technology cannot improve the diversity of images and cannot solve the generalization of various scenes

Images