A watermark embedding method based on a frequency domain space collaborative convolutional network

Abstract

The application discloses a watermark embedding method based on a frequency domain space collaborative convolution network and relates to the technical field of information security and computer vision. The method comprises the following steps: obtaining a carrier image and watermark information to be embedded; performing feature extraction on the carrier image to obtain image features; performing dimension expansion and space alignment on the watermark information to be embedded to obtain watermark features; inputting the image features and the watermark features into a pre-trained encoder after splicing, performing splicing and preliminary feature extraction on the image features and the watermark features through a two-dimensional convolution block to obtain joint features; performing space domain and frequency domain sensing feature extraction and fusion on the joint features through a frequency domain space collaborative convolution network in the encoder to obtain mixed domain features; performing attention weight prediction on the image features through an attention module in the encoder to obtain an attention mask; fusing the mixed domain features and the attention mask to obtain watermark disturbance features; and fusing the watermark disturbance features and the carrier image to obtain a watermark embedding image.

Images