Pixel

A digital image is built from tiny picture elements called pixels. Think of a pixel as one small measurement of light and colour at a particular position in an image; millions of these measurements together form the photos, video frames, scans, and camera feeds that a computer-vision system examines.

What a pixel stores
A pixel has a location, commonly written as coordinates (x, y), and one or more numeric values. In a grayscale image, one value represents brightness: lower numbers are darker and higher numbers are lighter. In a standard colour image, a pixel usually has three channels: red, green, and blue (RGB). With 8 bits per channel, each channel ranges from 0 to 255, so a pixel such as (255, 0, 0) is bright red. Libraries such as OpenCV store colour images in BGR order by default, a small but important detail when displaying or processing images.

Pixels as model input
Computer-vision pipelines turn an image into an array of pixel values. A 1920 × 1080 image contains more than two million pixel locations. Models use patterns across neighbouring pixels—not isolated values—to recognize edges, textures, shapes, and objects. For example:

• An OCR system examines pixel patterns to distinguish “8” from “B”.
• A medical segmentation model assigns each pixel a label such as tumour, organ, or background.
• A production-line inspection system finds unusually dark, bright, or irregular pixel regions that signal a defect.

Why pixels matter
Pixel values determine what visual evidence a model can use. Image resolution controls how much spatial detail is available: a tiny distant pedestrian may occupy too few pixels for an autonomous-vehicle detector to recognize reliably. Pixel-level preprocessing—resizing, normalizing values, correcting colour, or reducing noise—also directly affects model predictions. In tasks such as semantic segmentation, the desired output is itself a pixel-by-pixel map, making the pixel both the raw input unit and the unit in which the final visual decision is expressed.