The module brings implementations of different image hashing algorithms.

Detailed Description

Provide algorithms to extract the hash of images and fast way to figure out most similar images in huge data set.

Namespace for all functions is cv::img_hash.

Supported Algorithms

• Average hash (also called Different hash)
• PHash (also called Perceptual hash)
• Marr Hildreth Hash
• Radial Variance Hash
• Block Mean Hash (modes 0 and 1)
• Color Moment Hash (this is the one and only hash algorithm resist to rotation attack(-90~90 degree))

You can study more about image hashing from following paper and websites:

• "Implementation and benchmarking of perceptual image hash functions" [315]
• "Looks Like It" [150]

Code Example

#include "opencv2/core.hpp"
#include "opencv2/core/ocl.hpp"
#include "opencv2/highgui.hpp"
#include "opencv2/img_hash.hpp"
 
#include <iostream>
 
using namespace cv;
using namespace cv::img_hash;
using namespace std;
 
template <typename T>
inline void test_one(const std::string &title, const Mat &a, const Mat &b)
{
    cout << "=== " << title << " ===" << endl;
    TickMeter tick;
    Mat hashA, hashB;
    Ptr<ImgHashBase> func;
    func = T::create();
 
    tick.reset(); tick.start();
    func->compute(a, hashA);
    tick.stop();
    cout << "compute1: " << tick.getTimeMilli() << " ms" << endl;
 
    tick.reset(); tick.start();
    func->compute(b, hashB);
    tick.stop();
    cout << "compute2: " << tick.getTimeMilli() << " ms" << endl;
 
    cout << "compare: " << func->compare(hashA, hashB) << endl << endl;;
}
 
int main(int argc, char **argv)
{
    if (argc != 3)
    {
        cerr << "must input the path of input image and target image. ex : hash_samples lena.jpg lena2.jpg" << endl;
        return -1;
    }
    ocl::setUseOpenCL(false);
 
    Mat input = imread(argv[1]);
    Mat target = imread(argv[2]);
 
    test_one<AverageHash>("AverageHash", input, target);
    test_one<PHash>("PHash", input, target);
    test_one<MarrHildrethHash>("MarrHildrethHash", input, target);
    test_one<RadialVarianceHash>("RadialVarianceHash", input, target);
    test_one<BlockMeanHash>("BlockMeanHash", input, target);
 
    return 0;
}

Performance under different attacks

Performance chart

Speed comparison with PHash library (100 images from ukbench)

As you can see, hash computation speed of img_hash module outperform PHash library a lot.

PS : I do not list out the comparison of Average hash, PHash and Color Moment hash, because I cannot find them in PHash.

Motivation

Collects useful image hash algorithms into opencv, so we do not need to rewrite them by ourselves again and again or rely on another 3rd party library(ex : PHash library). BOVW or correlation matching are good and robust, but they are very slow compare with image hash, if you need to deal with large scale CBIR(content based image retrieval) problem, image hash is a more reasonable solution.

More info

You can learn more about img_hash modules from following links, these links show you how to find similar image from ukbench dataset, provide thorough benchmark of different attacks(contrast, blur, noise(gaussion,pepper and salt), jpeg compression, watermark, resize).

Introduction to image hash module of opencv Speed up image hashing of opencv(img_hash) and introduce color moment hash

Contributors

Tham Ngap Wei, thamn.nosp@m.gapw.nosp@m.ei@gm.nosp@m.ail..nosp@m.com

Classes
class	cv::img_hash::AverageHash
	Computes average hash value of the input image. More...
class	cv::img_hash::BlockMeanHash
	Image hash based on block mean. More...
class	cv::img_hash::ColorMomentHash
	Image hash based on color moments. More...
class	cv::img_hash::ImgHashBase
	The base class for image hash algorithms. More...
class	cv::img_hash::MarrHildrethHash
	Marr-Hildreth Operator Based Hash, slowest but more discriminative. More...
class	cv::img_hash::PHash
	pHash More...
class	cv::img_hash::RadialVarianceHash
	Image hash based on Radon transform. More...

Enumerations
enum	cv::img_hash::BlockMeanHashMode {   cv::img_hash::BLOCK_MEAN_HASH_MODE_0 = 0 ,   cv::img_hash::BLOCK_MEAN_HASH_MODE_1 = 1 }

Functions
void	cv::img_hash::averageHash (cv::InputArray inputArr, cv::OutputArray outputArr)
	Calculates img_hash::AverageHash in one call.
void	cv::img_hash::blockMeanHash (cv::InputArray inputArr, cv::OutputArray outputArr, int mode=BLOCK_MEAN_HASH_MODE_0)
	Computes block mean hash of the input image.
void	cv::img_hash::colorMomentHash (cv::InputArray inputArr, cv::OutputArray outputArr)
	Computes color moment hash of the input, the algorithm is come from the paper "Perceptual Hashing for Color Images Using Invariant Moments".
void	cv::img_hash::marrHildrethHash (cv::InputArray inputArr, cv::OutputArray outputArr, float alpha=2.0f, float scale=1.0f)
	Computes average hash value of the input image.
void	cv::img_hash::pHash (cv::InputArray inputArr, cv::OutputArray outputArr)
	Computes pHash value of the input image.
void	cv::img_hash::radialVarianceHash (cv::InputArray inputArr, cv::OutputArray outputArr, double sigma=1, int numOfAngleLine=180)
	Computes radial variance hash of the input image.

Enumeration Type Documentation

BlockMeanHashMode

enum cv::img_hash::BlockMeanHashMode

#include <opencv2/img_hash/block_mean_hash.hpp>

Enumerator
BLOCK_MEAN_HASH_MODE_0  Python: cv.img_hash.BLOCK_MEAN_HASH_MODE_0	use fewer block and generate 16*16/8 uchar hash value
BLOCK_MEAN_HASH_MODE_1  Python: cv.img_hash.BLOCK_MEAN_HASH_MODE_1	use block blocks(step sizes/2), generate 31*31/8 + 1 uchar hash value

Function Documentation

averageHash()

void cv::img_hash::averageHash	(	cv::InputArray	inputArr,
		cv::OutputArray	outputArr
	)

Python:
	cv.img_hash.averageHash(	inputArr[, outputArr]	) ->	outputArr

#include <opencv2/img_hash/average_hash.hpp>

Calculates img_hash::AverageHash in one call.

Parameters

inputArr	input image want to compute hash value, type should be CV_8UC4, CV_8UC3 or CV_8UC1.
outputArr	Hash value of input, it will contain 16 hex decimal number, return type is CV_8U

blockMeanHash()

void cv::img_hash::blockMeanHash	(	cv::InputArray	inputArr,
		cv::OutputArray	outputArr,
		int	mode = BLOCK_MEAN_HASH_MODE_0
	)

Python:
	cv.img_hash.blockMeanHash(	inputArr[, outputArr[, mode]]	) ->	outputArr

#include <opencv2/img_hash/block_mean_hash.hpp>

Computes block mean hash of the input image.

Parameters

inputArr	input image want to compute hash value, type should be CV_8UC4, CV_8UC3 or CV_8UC1.
outputArr	Hash value of input, it will contain 16 hex decimal number, return type is CV_8U
mode	the mode

colorMomentHash()

void cv::img_hash::colorMomentHash	(	cv::InputArray	inputArr,
		cv::OutputArray	outputArr
	)

Python:
	cv.img_hash.colorMomentHash(	inputArr[, outputArr]	) ->	outputArr

#include <opencv2/img_hash/color_moment_hash.hpp>

Computes color moment hash of the input, the algorithm is come from the paper "Perceptual Hashing for Color Images Using Invariant Moments".

Parameters

inputArr	input image want to compute hash value, type should be CV_8UC4, CV_8UC3 or CV_8UC1.
outputArr	42 hash values with type CV_64F(double)

marrHildrethHash()

void cv::img_hash::marrHildrethHash	(	cv::InputArray	inputArr,
		cv::OutputArray	outputArr,
		float	alpha = 2.0f,
		float	scale = 1.0f
	)

Python:
	cv.img_hash.marrHildrethHash(	inputArr[, outputArr[, alpha[, scale]]]	) ->	outputArr

#include <opencv2/img_hash/marr_hildreth_hash.hpp>

Computes average hash value of the input image.

Parameters

inputArr	input image want to compute hash value, type should be CV_8UC4, CV_8UC3, CV_8UC1.
outputArr	Hash value of input, it will contain 16 hex decimal number, return type is CV_8U
alpha	int scale factor for marr wavelet (default=2).
scale	int level of scale factor (default = 1)

pHash()

void cv::img_hash::pHash	(	cv::InputArray	inputArr,
		cv::OutputArray	outputArr
	)

Python:
	cv.img_hash.pHash(	inputArr[, outputArr]	) ->	outputArr

#include <opencv2/img_hash/phash.hpp>

Computes pHash value of the input image.

Parameters

inputArr	input image want to compute hash value, type should be CV_8UC4, CV_8UC3, CV_8UC1.
outputArr	Hash value of input, it will contain 8 uchar value

radialVarianceHash()

void cv::img_hash::radialVarianceHash	(	cv::InputArray	inputArr,
		cv::OutputArray	outputArr,
		double	sigma = 1,
		int	numOfAngleLine = 180
	)

Python:
	cv.img_hash.radialVarianceHash(	inputArr[, outputArr[, sigma[, numOfAngleLine]]]	) ->	outputArr

#include <opencv2/img_hash/radial_variance_hash.hpp>

Computes radial variance hash of the input image.

Parameters

inputArr	input image want to compute hash value, type should be CV_8UC4, CV_8UC3, CV_8UC1.
outputArr	Hash value of input
sigma	Gaussian kernel standard deviation
numOfAngleLine	The number of angles to consider