U-Net++ Mosaic Network for Blue and White Porcelain Fragment Images

Aiming at the problems of low accuracy and robustness caused by artifacts at the stitching and content distortion in non-overlapping areas in the existing image stitching methods, a blue and white porcelain fragment image stitching method based on U-Net ++ to eliminate artifacts was proposed. Firstly, the homography matrix of the image to be concatenated is estimated, and then the homography matrix is applied to the structure concatenation stage to obtain the rough concatenation result of the image. Finally, the rough concatenation result of the image is used as the prior information, and the existing U-Net is improved in the content correction stage, and the rough concatenation result is refined by U-Net++ to obtain the final image accurate concatenation. The experimental results of the blue and white porcelain fragment image dataset and related classical methods show that, among the three evaluation indicators, the peak signal to noise ratio of the proposed method is increased by about 13%, the root mean square error is reduced by about 33%, and the mean square error is reduced by about 57%. This method can not only improve the quality of image mosaic, but also show good robustness with small error ratio.