ABSTRACT
This chapter presents a new shape descriptor suitable for machine contour perception that mimics the orientation selective cells of the human visual cortex. It provides a description for any 2-dimensional shape, regardless of its size, rotation and position, with affordable computational cost. Furthermore, it can identify different shapes of the same object, distorted shapes and heavily contaminated shapes with up to 20% spike noise. The earliest stage of visual processing in the Human Visual System (HVS) is located in the primary visual cortex in an area known as VI. Similarly to the HVS, the proposed shape descriptor uses 12 groups of artificial VI cells each one specialized in detecting any edge with a particular orientation within its receptive field. Cells from higher levels of the HVS, receive whole visual maps of stimulated VI cells, and identify certain combinations among them, resulting to the gradual perception of more complex forms.
