A real scene image synthesis method and system

Abstract

The invention discloses a real scene image synthesis method and system. The method comprises the following steps of obtaining an image training set composed of a semantic map and a real scene reference map corresponding to the semantic map; according to U-NET convolution neural network model and an excitation residual block, establishing the real scene image synthesis network model; using pre-trained VGG-19 convolution neural network model to establish the loss function of real scene image synthesis network model; using the image training set as the input of the real scene image synthesis network model; according to the loss function, training the real scene image synthesis network model to obtain the trained real scene image synthesis network model; acquiring a plurality of semantic mapsto be synthesized; inputting the semantic image to be synthesized into the trained real scene image synthesis network model, and obtaining the real scene composite image corresponding to the semanticimage to be synthesized. The invention can quickly and effectively synthesize the photographic-level scene image with stronger sense of reality, and improves the sense of reality and the visual quality of the synthesized image.

Description

technical field [0001] The present invention relates to the technical field of image synthesis, in particular to a real scene image synthesis method and system. Background technique [0002] In the field of image synthesis, real scene image synthesis technology based on deep learning is gradually being applied more and more widely. Real-scene image synthesis is a visual image synthesis technology that synthesizes images close to real scenes based on the object segmentation information in the semantic layout graph. Real-scene image synthesis methods integrate a variety of professional technologies such as deep learning, pattern recognition, and digital image processing. There are three key points in real scene image synthesis: (1) global coordination; (2) storage capacity of network model; (3) high resolution. Deep learning can realize global image feature extraction, and can also increase the number of parameters of the network model, that is, the storage capacity of the n...

AI Technical Summary

Problems solved by technology

Although CRNs have huge storage capacity, it can generate more realistic images than method (1), but it takes a lot of time in the training and prediction stages, and cannot achieve fast and efficient synthesis of real scene images

In short, the efficiency of existing synthesis methods is low, and the realism and visual quality of the synthesized photorealistic images need to be improved

Images