**4. Results and Discussion**

After the feature extraction by the deep CNN, each image was assigned with HDFV and its corresponding attributes (semantic information and infrastructure information). Four cases were selected to explain the HDFV performance in representing the morphological similarity of the images. The urban fabric images and the distances between each HDFV are listed in Figure 8. We conclude that plots b and c, and a and d are pairs of similar morphological types according to the HDFV distances. Plots a and d are distributed with intensively lined-up buildings. These are aged residential areas built in the 20th century. Plots b and c were built in recent decades, also with lined-up buildings but with more sparse textures.


**Figure 8.** The Euclidean distance between the samples.
