Increasing the Size of the Camouflage Area for Remote Sensing Images

Image camouflaging is a hard-to-find replacement of some information in an image with a verisimilar imitation. An increase in the size of the camouflage area entails a degradation in the quality of the results of most camouflage algorithms. The article deals with the effect of the camouflaged area size on the performance of the adversarial neural network when camouflaging remote sensing images. We consider a camouflage adversarial neural network that includes an algorithm for edge analysis and generation. We study the dependence of the qualitative and quantitative characteristics of remote sensing camouflage images on the size of the camouflaged area. We use a quantitative measure of quality based on the image sharpness inside the camouflage area. We compute an estimate of the image sharpness within the camouflage area using a gradient averaged over that area. We average this gradient value over a series of natural camouflage images in our experimental study. Computational experiments prove a high quality of the adversarial network result with an increase in the size of the camouflage areas for natural remote sensing images.