ISCC Primer¶

What is ISCC?¶

ISCC (International Standard Content Code, ISO 24138) is an open standard for identifying digital content. Unlike traditional identifiers that are assigned by a registry, ISCC codes are generated directly from the content itself. Think of it as a DNA test for digital files: two people processing the same content independently will produce the same ISCC.

ISCC codes are similarity-preserving. Small changes to content produce small changes in the code. This property makes ISCC useful for finding near-duplicates, derivative works, and related content across large collections.

ISCC-CODE vs ISCC-ID¶

ISCC has two complementary identifier types that serve different purposes.

ISCC-CODE is a deterministic fingerprint generated from content. You feed a file (text, image, audio, video) into the ISCC algorithm and get back a code that describes what the content looks like, sounds like, and contains. ISCC-CODEs are reproducible: the same input always produces the same code. They answer the question "what is this content?"

ISCC-ID is a persistent identifier issued by a registration network. It records who registered the content, when, and where. ISCC-IDs are unique and never change once issued. They answer the question "who registered this content?"

An analogy: ISCC-CODE is like a fingerprint - derived from what you are. ISCC-ID is like a passport number - assigned by an authority.

ISCC-UNITs¶

An ISCC-CODE is composed of multiple ISCC-UNITs, each capturing a different aspect of the content.

graph TB
    CODE[<b>ISCC-CODE</b><br/>Multi-component Fingerprint]
    META[<b>META-UNIT</b><br/><i>title, description</i>]
    SEMANTIC[<b>SEMANTIC-UNIT</b><br/><i>semantic meaning</i>]
    CONTENT[<b>CONTENT-UNIT</b><br/><i>perceptual features</i>]
    DATA[<b>DATA-UNIT</b><br/><i>raw data structure</i>]
    INSTANCE[<b>INSTANCE-UNIT</b><br/><i>exact identity</i>]
    CODE -->|composed of| META & SEMANTIC & CONTENT & DATA & INSTANCE

Each unit type serves a specific purpose:

• META-UNIT captures similarity in title, creator, and description text. Two works with similar titles produce similar META-UNITs.
• SEMANTIC-UNIT captures the conceptual meaning of content using AI embeddings. A text and its translation into another language share a similar SEMANTIC-UNIT.
• CONTENT-UNIT captures perceptual features - how content looks or sounds to humans. A cropped photo and the original share a similar CONTENT-UNIT.
• DATA-UNIT captures the raw bitstream structure. Re-encoded versions of the same file (e.g., JPEG to PNG) share a similar DATA-UNIT.
• INSTANCE-UNIT acts as a cryptographic checksum for exact identity. Two byte-identical files produce the same INSTANCE-UNIT.

Not every ISCC-CODE contains all unit types. DATA and INSTANCE are mandatory. The others depend on what information is available during generation.

ISCC-SIMPRINTs¶

ISCC-SIMPRINTs are granular, segment-level fingerprints. While an ISCC-CODE describes an entire asset, SIMPRINTs describe individual chunks within the asset. One document or audio file produces many SIMPRINTs, one per segment.

SIMPRINTs enable partial matching. You can find where a specific paragraph appears within a larger book, or detect that a 30-second audio clip was extracted from a full song. The search operates at the chunk level and aggregates results back to the asset level.

Variable-length codes¶

ISCC codes come in different bit lengths: 64, 128, 192, or 256 bits. The length determines the precision of matching:

• 64-bit codes are compact and fast to compare. They provide broad matching that catches more candidates but with more false positives.
• 256-bit codes are more precise. They reduce false positives but may miss content with larger modifications.

A key property of ISCC: shorter codes are valid prefixes of longer ones. A 64-bit code is the first 64 bits of the 128-bit code for the same content. This means you can compare codes of different lengths by matching on their common prefix. The iscc-search engine handles this automatically using the NPHD metric.

The reverse discovery problem¶

Traditional identifiers work in one direction: you know the ID and look up the content. ISBN 978-3-16-148410-0 points to a specific book. But if you have the book and want to find its ISBN, you need a separate lookup service.

ISCC solves this content-to-identifier problem. You can fingerprint any digital file and search for it in an ISCC index. If someone registered that content (or similar content), you find it. No prior knowledge of the identifier is needed.

This is what iscc-search provides: you submit content or an ISCC code, and the engine finds registered assets that match at various similarity levels - from exact duplicates to loosely related works.

Now that you understand ISCC concepts, see Getting started to try it hands-on.