Error level analysis in Image - opencv

Question

How do I compute ELA for an image? I would like to get similar ELA image using opencv http://fotoforensics.com/tutorial-ela.php
As per this tutorial, I resaved the image at 95% quality jpeg image and using absDiff method to compute the difference between the source image and the resaved image but all I am getting is zero difference.
Any help on how to compute the difference between two images so as to obtain the error level just like sample images in the tutorial?

Answer 1

The key to achieve a similar result is to use a variable value for the compression rate and a scale factor to make it easier to visualize the data.
Here's an example: we have the input image (left) and the processed image after some parameter adjustments (right):
As expected, the region with the christmas hat presents a different compression rate from the rest of the image. This result is very similar to what FotoForensics presents:
With a few tweaks on this code you can achieve an even closer result. The source code of this project can be found on my Github:
main.cpp:
#include <opencv2/highgui/highgui.hpp>
#include <iostream>
#include <vector>
// Control
int scale = 15,
quality = 75;
// Image containers
cv::Mat input_image,
compressed_image;
void processImage(int, void*)
{
// Setting up parameters and JPEG compression
std::vector<int> parameters;
parameters.push_back(CV_IMWRITE_JPEG_QUALITY);
parameters.push_back(quality);
cv::imwrite("temp.jpg", input_image, parameters);
// Reading temp image from the disk
compressed_image = cv::imread("temp.jpg");
if (compressed_image.empty())
{
std::cout << "> Error loading temp image" << std::endl;
exit(EXIT_FAILURE);
}
cv::Mat output_image = cv::Mat::zeros(input_image.size(), CV_8UC3);
// Compare values through matrices
for (int row = 0; row < input_image.rows; ++row)
{
const uchar* ptr_input = input_image.ptr<uchar>(row);
const uchar* ptr_compressed = compressed_image.ptr<uchar>(row);
uchar* ptr_out = output_image.ptr<uchar>(row);
for (int column = 0; column < input_image.cols; column++)
{
// Calc abs diff for each color channel multiplying by a scale factor
ptr_out[0] = abs(ptr_input[0] - ptr_compressed[0]) * scale;
ptr_out[1] = abs(ptr_input[1] - ptr_compressed[1]) * scale;
ptr_out[2] = abs(ptr_input[2] - ptr_compressed[2]) * scale;
ptr_input += 3;
ptr_compressed += 3;
ptr_out += 3;
}
}
// Shows processed image
cv::imshow("Error Level Analysis", output_image);
}
int main (int argc, char* argv[])
{
// Verifica se o número de parâmetros necessário foi informado
if (argc < 2)
{
std::cout << "> You need to provide an image as parameter" << std::endl;
return EXIT_FAILURE;
}
// Read the image
input_image = cv::imread(argv[1]);
// Check image load
if (input_image.empty())
{
std::cout << "> Error loading input image" << std::endl;
return EXIT_FAILURE;
}
// Set up window and trackbar
cv::namedWindow("Error Level Analysis", CV_WINDOW_AUTOSIZE);
cv::imshow("Error Level Analysis", input_image);
cv::createTrackbar("Scale", "Error Level Analysis", &scale, 100, processImage);
cv::createTrackbar("Quality", "Error Level Analysis", &quality, 100, processImage);
// Press 'q' to quit
while (char(cv::waitKey(0)) != 'q') {};
return EXIT_SUCCESS;
}
Here are some nice references that were used to build this mash-up:
ELA with HTML5
FotoForensics Tutorial
Blackhat USA '07 Paper