Obtaining missing areas with exemplar inpainting in mosaic unmanned aerial vehicle images

In this article, mosaic process was applied to use in digital elevation model of terrestrial images by using Unmanned Aerial Vehicle (UAV). Images acquired with an industrial UAV with a compact camera, rotary-wing, at 120 m flight height (similar to 3.3 cm/pixel Ground Sampling Range -GSD-) 80% frontal overlap and 50% side overlap. The images were combined using 50 Ground Control Points (GCP) established in the test area at Konya Selcuk University Campus Area. The orthomosaic obtained from the images produced by Pix4D software. A total of 173 UAV images used for the mosaicing process. The performance of the proposed exemplar inpainting method was tested on 6228 images in size of 256 x 256 which genereated from 173 UAV images and the orthomosaic image generated with 173 UAV images. In the proposed method, different image patch sizes of 5 x 5, 7 x 7 and 9 x 9 are used for inpainting process. The performance of the proposed method according to different patch sizes was evaluated. Structural Similarity Index (SSIM) was obtained as 0.9824 for 5 x 5 patch size, 0.9840 for 7 x 7 patch size and 0.9843 for 9 x 9 patch size. Signal to Noise Ratio (SNR) was obtained as 22.1010 dB for 5 x 5 patch size, 22.5148 dB for 7 x 7 patch size and 22.6927 dB for 9 x 9 patch size. Peak Signal Noise Ratio (PSNR) was 21.7303 dB for the 5 x 5 patch size, 21.3184 dB for the 7 x 7 patch size and 21.1420 dB for the 9 x 9 patch size. Finally, inpainting was performed on missing areas in the orthomosaic image by using proposed method.