Video narration
How does your phone turn pixels into "cat" without seeing the whole picture like you do? A photo starts as numbers: each pixel records visual information. A convolutional neural network, or CNN, examines nearby pixels at a time. Its key tool is a filter-a small learned pattern detector-that slides across the image, asking where a simple pattern appears. At first, those patterns are often edges. Deeper layers combine them into curves, textures, and shapes, then clues about animal parts. Finally, it combines those clues into a prediction: cat, with some confidence. That is the turn: not the whole image at once. Recognition builds from tiny learned patterns. And it does not learn to see like a person. During training, examples adjust the filters so useful patterns become easier to detect. So how does it recognize the cat? It combines edges, shapes, and parts until the photo fits.