In fact, machines are not good at recognizing images. For example, this picture is just a string of data consisting of 0 and 1 in the eyes of the machine, and the machine can't understand the meaning of this picture. So if we want the machine to learn to recognize images, we need to write a program algorithm for it.
When we describe a person's appearance, we usually use words like melon seeds, willow eyes, garlic nose and cherry mouth. The so-called appearance, to a large extent, depends on people's head shape and facial features.
This method was adopted in the earliest face recognition. First, the machine will identify the position of the face in the image, and then draw the outline of the five senses on this face, thus obtaining the shape and position information of the five senses on the face. For example, the distance between the eyes, the connecting angle between the tip of the nose and the corner of the mouth in the horizontal direction and so on.