Skip to main content
. 2018 Apr 17;18(4):1232. doi: 10.3390/s18041232

Figure 1.

Figure 1

Reconstruction using an autoencoder. We trained the autoencoder using patch images as reference images. We cropped patch images for a current image in (a) and reconstructed each patch in (b). It is difficult to select a threshold value to distinguish the foreground clearly using the differences, as shown in (c).