A pixel is the smallest addressable picture element in a digital image. In an RGB image, each pixel is represented by three numbers between 0 and 1, describing its red, green, and blue components. An image can therefore be treated as a vector of numerical pixel values for machine-learning systems. section 3.4
Recognizing cats and dogs
A deep-learning system classifying photographs of cats and dogs receives images as collections of pixel values. It learns patterns hierarchically, moving from shapes toward colors, textures, and finer features. Reliable recognition requires many high-resolution, human-labeled examples showing different angles, activities, sizes, and appearances. section 3.4 section 6.2
Pixel-based representations support visual learning but impose important limits:
- Pixels provide numerical signals from which models can identify visual patterns. section 3.4
- Pixel values do not by themselves give a machine the meaning of an image or the possible uses of an object. section 6.2
