Figure 4

Patch distribution at image level for images obtained from unsupervised deep learning applied on dried droplets of Viscum album Quercus 3× produced by means of diffusion-based mixing (variant D) (a), turbulent mixing (variant T) (b), and laminar flow induced by a handmade vortex (variant L) (c). The bias toward more and less fractal behavior of the dried droplet is visible for the laminar and turbulent modalities, while the diffusion modality presents both biases.