An Improved SURF in Image Mosaic Based on Deep Learning
Hao Lv, Haiyang Zhang, Changming Zhao, Chun Liu, QI Faguo, Zilong Zhang · 2019
Image mosaic is widely used in the machine vision and various projects, the mosaic image may sometimes in malformed, which is impossible to be predicted when it occurs, with the Speed-up Robust Feature (SURF). In order to get successful mosaic images at most of the time with SURF, an improved SURF based on deep learning with Convolutional Neural Network (CNN) is proposed in this paper. The improved SURF can feedback the mosaic procession by re-stitching the images automatically after eliminating the bad deformed one by the judging of the results through deep learning, and output the better, which free-data area of black will be continue to be detected and then be restored with the nearest neighbor interpolation, and finally outputs the ideal image. The experiment shows that the improved SURF can eliminate the deformed mosaic as effectively and can improve the mosaic as efficiency, and get the final perfect mosaic images all the time.