Method for separating two mutually exclusive components in image

Abstract

The invention discloses a method for separating two mutually exclusive components in an image, the method is based on a neural network of a double-flow information interaction module, and the method is characterized in that the double-flow information interaction module adopts the following steps to separate the two mutually exclusive components in the image: S1, two parallel convolutional layers in the neural network output image features of an acquired image; S2, a second decomposition unit adopts two branches to activate the image features to obtain first activated image features and second activated image features; and S3, the fusion image unit combines the first activation image feature and the second activation image feature to generate a first combination feature and a second combination feature. Compared with a traditional double-branch network which is activated by using a ReLU activation function, the method has the advantages that the negative ReLU activation function is introduced to retain features abandoned by the ReLU activation function, and the features are transmitted to the other branch, so that the total amount of information is maintained, information loss is avoided, and the transmission efficiency of the information is improved.

Description

technical field [0001] The invention belongs to image processing methods, in particular to a method for separating two mutually exclusive components in an image. Background technique [0002] The existing general-purpose dual-branch image restoration framework is a general-purpose image restoration network structure that can be used for image compression artifact removal, image denoising, and image super-resolution, and the dual-branch strategy in this work decomposes image restoration into texture and structure Two branches, and use a convolutional layer as the interaction between the two branches. But in fact, this work only includes one-way interaction (Lateral Connections), and does not filter the characteristics of the interaction, only uses convolution to perform affine transformation, and the efficiency of information utilization is poor. [0003] Another dual-branch image restoration structure is used to deal with a variety of composite image degradation problems. ...

AI Technical Summary

Problems solved by technology

In addition, this mode cannot be generalized to other tasks. For example, in the image reflection decomposition task, this summing method will cause the feature content of the two branches to be almost the same, which is not conducive to learning mutually exclusive component features.

Images