"Image search" is not one technique — it's a family of them, each suited to a different question. Are you trying to find where a photo came from? Identify an object in it? Check whether two images are secretly the same? Read text baked into a screenshot? Each of those is a different technique under the hood, and knowing which one to reach for saves a lot of wasted effort.
Keyword-based search
The oldest form of image search: you type words and a search engine matches them against filenames, alt text, captions, and surrounding page content. It's fast and familiar, but it only works if someone has already described the image in text — a photo with no caption or a generic filename like IMG_4821.jpg is effectively invisible to keyword search.
Reverse image search
Instead of describing an image in words, you submit the image itself and ask 'where else does this appear, and what is it?' Engines like Google Lens, TinEye, Yandex, and Bing Visual Search analyze the image's visual features and compare them against indexes of billions of crawled images. This is the technique of choice for finding the original source of a photo, spotting stolen or reused images, or identifying an object, landmark, or product you don't have words for.
Visual similarity (perceptual hashing)
Rather than an exact byte-for-byte match, perceptual hashing algorithms (like pHash and dHash) reduce an image to a compact fingerprint based on its visual structure — gradients, luminance patterns, edges. Two images with similar fingerprints look alike to the human eye, even if one has been resized, recompressed, or lightly cropped. This is how many duplicate-detection and content-moderation systems work under the hood, and it's the technique behind our own image similarity tool below.
Metadata-based search
Every photo taken on a modern camera or phone typically carries an EXIF data block: camera make and model, exposure settings, timestamp, and — if location services were on — GPS coordinates. Metadata search means filtering or discovering images based on this embedded data rather than pixel content. It's powerful for organizing personal photo libraries by date or location, and it's also a privacy consideration: metadata you don't strip travels with the file wherever it's shared.
Color-based search
Color-based techniques index images by their dominant or average color palette, letting you search for 'images that are mostly teal and orange' the way a designer might browse a mood board. It's less about identifying what an image is and more about how it feels visually — useful for stock photo libraries, design systems, and matching brand palettes.
OCR-based search
Optical character recognition treats the text inside an image — a screenshot, a scanned document, a photo of a sign — as searchable content. Instead of matching pixels or metadata, OCR extracts the literal words in the image so they can be copied, indexed, or searched like any other text. It's the bridge between 'this is a picture' and 'this is a document.'