The visual cortex, which is intended to solve image recognition problems, shows a sequence of sectors placed in a hierarchy. Each of these areas receives an input representation, by means of flow signals that connect it to other sectors.
Each level of this hierarchy represents a different level of abstraction, with the most abstract features defined in terms of those of the lower level. At a time when the brain receives an input image, the processing goes through various phases, for example, detection of the edges or the perception of forms (from those primitive to those gradually more and more complex).
As the brain learns by trial and activates new neurons by learning from the experience, even in deep learning architectures, the extraction stages or layers are changed based on the information received at the input.
The scheme, on the next page shows what has been said in the case of an image classification system, each block gradually extracts the features of the input image, going on to process data already preprocessed from the previous blocks, extracting features of the image that are increasingly abstract, and thus building the hierarchical representation of data that comes with on deep learning based system.
More precisely, it builds the layers as follows along with the figure representation:
- Layer 1: The system starts identifying the dark and light pixels
- Layer 2: The system identifies edges and shapes
- Layer 3: The system learns more complex shapes and objects
- Layer 4: The system learns which objects define a human face
Here is the visual representation of the process:
Figure 2: A deep learning system at work on a facial classification problem