canvas-phash

Note: This project is no longer actively maintained. The following README content needs to be updated (as it's been 8 years since this project was revised) but the general ideas hold. I've removed bluebird as a dependency, which might break backwards compatibility for some use-cases. If you're updating to v3 from v2, that should be the only breaking change.

Introduction

This is an implementation of a perceptual image hash, using Canvas written in 100% javascript/coffeescript. The algorithm used is described in Block Mean Value Based Image Perceptual Hashing and discussed in this StackOverflow question.

Difference From phash

I found the phash package to be a little error prone with respect to file I/O. This package, while the API is very similar, is different in some key ways.

• phash binds directly to the pHash library. canvas-phash is a direct implementation, written in coffeescript.
• phash is callback-based while canvas-phash is promise-based (specifically, it uses bluebird for promise management).
• phash generally takes longer to compute the hash of an image but is faster at finding the hamming distance between two hashes.
• The hash output by phash is an integer, expressed as a string. The hash output by canvas-phash is a 128-byte Buffer.
• Comparing the two libraries on the basis of the correlation between hamming distance and "perceived difference" had mixed results. phash was better at some things, canvas-phash was better at others.

Performance

I ran some preliminary tests to check the performance against phash and found it's fairly comparable.

Computing A Hash

The time taken ranged from just under 75ms to 150ms. For my tests, it generally took phash about 1-2 times longer to compute a hash as it took canvas-phash.

Finding the Hamming Distance

Typical time taken ranged from 0.2ms to 0.3ms. For my tests, it generally took canvas-phash about 2-3 times longer to find the hamming distance of two hashes. When comparing against a large collection of images, this is potentially significant. That being said, this library has not been optimized. Also, the actual hash created is 128 bytes long and takes up about 2-3 times more space.

API

• getImageHash - Accepts 1 parameter, the path of the image. Returns a promise with eventual value equal to the "Block Mean Value Based" pHash.
• getHammingDistance - Accepts 2 parameters, two instances of Buffer of length 128 (this is what is returned from getImageHash)
• getSHA256 - This computes the SHA256 hash of the pixel data. The only parameter is setup like that of getImageHash. This is useful for fast checks of exact matches. Ignores metadata.
• readImage - Reads an image at the specified path and returns an object with properties: data, the byte array, width, the width of the image, and height, the height of the image.

Example Usage

(Another example exists in the repo)

phash = require 'canvas-phash'

Promise = require 'bluebird'
Promise.all([
	phash.getImageHash 'image.jpg'
	phash.getImageHash 'otherImage.jpg'
])
.spread (hash1, hash2)->
	dist = phash.getHammingDistance hash1, hash2

In the previous example, Promise.all is used to make the code readable. requireing bluebird is not necessary to use this package. The typical use-case would be to compute the hash of a single image via phash.getImageHash('image.jpg').then (hash)-> and compare that against a list of pre-existing hashes for close matches.