Video4j - Fingerprint

This project contains a Digital video fingerprinting implementation.

Details on the actual fingerprinting process can be found in this dedicated Blog post.

The actual fingerprinting process is relatively fast as it does not require complex transformation or edge detection. Downside of the used stacking approach is however that there is no robustness against image rotation or mirroring of the image. Other solutions like pHash might be better if this is a problematic limitation.

Maven

<dependency>
  <groupId>io.metaloom.video</groupId>
  <artifactId>video4j-fingerprint</artifactId>
  <version>1.2.0</version>
</dependency>

Usage

Video4j.init();

// Create a fingerprinter for the video
MultiSectorVideoFingerprinter gen = new MultiSectorVideoFingerprinterImpl();

// Open the video using the Video4j API
try (Video video = Videos.open("src/test/resources/Big_Buck_Bunny_720_10s_30MB.mp4")) {

  // Run the actual hashing process
  MultiSectorFingerprint fingerprint = gen.hash(video);
  String hex = fingerprint.hex();
  // hex = 0001000100ff060006000f002e001d0084000600e40076d172c07c84ffcefffffefff8fffdff

  // Or get the binary form of the fingeprint
  byte[] bin = fingerprint.array();

  // Access the vector data
  float[] vec = fingerprint.vector();

  // Print information about the fingerprint data
  System.out.println(fingerprint.toString());

}