Perceptive

Perceptive is a Go library which implements perceptual hash algorithms for comparing images.

Perceptual hash algorithms are a family of comparable hash functions which generate distinct (but not unique) fingerprints, these fingerprints are then comparable.

Perceptual hash algorithms are mainly used for detecting duplicates of the same files, in a way that standard and cryptographic hashes generally fail.

Note: This library can only compute hashes for images, it does not work on audio or video files.

Currently, the following perceptual hash algorithms are implemented:

• Average Hash (Ahash) - Fast but generates a huge number of false positives.
• Difference Hash (Dhash) - Fast and very few false positives.

Perceptual hash algorithms can give false positives, but there main aim is to give you a sense of similarity between files.

Perceptual hash algorithms tend to return a distance score. When comparing the two identical images below, we would receive a distance of 0:

<img src="https://github.com/umahmood/perceptive/blob/master/perceptive_test/test_images/rainbow_flowers.jpg" width="310" height="300"/> <img src="https://github.com/umahmood/perceptive/blob/master/perceptive_test/test_images/rainbow_flowers.jpg" width="310" height="300"/>

A distance of zero means that the images are likely the same.

When comparing the two similar images below we would receive a distance between 1-10 (depending on the hashing technique used):

<img src="https://github.com/umahmood/perceptive/blob/master/perceptive_test/test_images/toy_story_1.jpg" /> <img src="https://github.com/umahmood/perceptive/blob/master/perceptive_test/test_images/toy_story_2.jpg" />

A distance between 1-10 indicates the images are likely a variation of each other.

When comparing the two different images below we would receive a distance greater than 10:

<img src="https://github.com/umahmood/perceptive/blob/master/perceptive_test/test_images/homer_doh.jpg" width="320" height="240"/> <img src="https://github.com/umahmood/perceptive/blob/master/perceptive_test/test_images/lena.jpg" width="320" height="240"/>

A distance greater than 10 indicates the images are likely different.

Remember perceptual hash algorithms can give false positives.

Installation

go get github.com/umahmood/perceptive

cd $GOPATH/src/github.com/umahmood/perceptive

go test ./...

Dependencies

• github.com/umahmood/iter
• github.com/disintegration/imaging

Usage

package main

import (
    "log"

    "github.com/disintegration/imaging"
    "github.com/umahmood/perceptive"
)

func openImage(filePath string) image.Image {
    img, err := imaging.Open(filePath)
    if err != nil {
        log.Fatalln(err)
    }
    return img
}

func main() {
    imgA := openImage("lena.jpg")
    imgB := openImage("lena.jpg")

    distance, err := perceptive.CompareImages(imgA, imgB, perceptive.Difference)

    if distance == 0 {
        // images are likely the same
    } else if distance >= 1 && distance <= 10 {
        // images are potentially a variation
    } else {
        // images are likely different
    }
}

To do

• Implement Phash algorithm targeting images.
• Compute perceptual hashes for audio files.
• Compute perceptual hashes for video files.

Documentation

http://godoc.org/github.com/umahmood/perceptive

References

• http://phash.org/
• http://www.hackerfactor.com/blog/index.php?/archives/529-Kind-of-Like-That.html

License

See the LICENSE file for license rights and limitations (MIT).