How to measure the image similarity for noisy image (bad image) from the other set of images (good images)?

I have a noisy image (blurred or with lots of zeros values ) and I have large set of reconstructed images (good images). I have to select the corresponding best out of the reconstructed images. How do I perform this operation efficiently? What measure will be best way to perform do it? Any suggestion to find the best approx image.

[Image Analyst]

What's your definition of "best"? And what does the one noisy image have at all to do with finding a "best" image from some other group of images?

[KSSV]

Few checks are psnr, ssim.

[DGM]

I guess another question to ask is whether you have a reference image to compare against. If you don't, then you'll have to figure out how to even define what good means.

[Amit Arya]

Hello Image Analyst, let me elaborate my question further, I have image patches that are blurred or noisy, i have the set of reconstructed patches(mostly noise free). But I don't know for the blurry or noisy image which one is the best match in reconstruced set of image patches. What measure should I consider or how do I perform search over the recostructed set of image patches to find the best match?

Hi DGM, yes I have the reference image to match the quality of the reconstruced image at the end for comparision but before that I have to find the best patches set from the set of reconstructed patches.

[DGM]

If you have a reference image, KSSV's suggestion may be of value.

[Amit Arya]

Can you please elaborate KSSV? with why and how?

Answer 1

[Image Analyst]

Try this. Adapt as needed:

clc;    % Clear the command window.
close all;  % Close all figures (except those of imtool.)
clear;  % Erase all existing variables. Or clearvars if you want.
workspace;  % Make sure the workspace panel is showing.
format long g;
format compact;
fontSize = 20;

% Specify the folder where the files live.
inputFolder = 'C:\Users\yourUserName\Documents\My Pictures';
outputFolder = 'C:\Users\yourUserName\Documents\My Pictures';
% Check to make sure that folder actually exists.  Warn user if it doesn't.
if ~isfolder(inputFolder)
	errorMessage = sprintf('Error: The following folder does not exist:\n%s\nPlease specify a new folder.', inputFolder);
	uiwait(warndlg(errorMessage));
	inputFolder = uigetdir(); % Ask for a new one.
	if inputFolder == 0
		% User clicked Cancel
		return;
	end
end
% Check to make sure that folder actually exists.  Warn user if it doesn't.
if ~isfolder(outputFolder)
	errorMessage = sprintf('Error: The following folder does not exist:\n%s\nPlease specify a new folder.', outputFolder);
	uiwait(warndlg(errorMessage));
	outputFolder = uigetdir(); % Ask for a new one.
	if outputFolder == 0
		% User clicked Cancel
		return;
	end
end
% Get a list of all files in the folder with the desired file name pattern.
filePattern = fullfile(inputFolder, '*.png'); % Change to whatever pattern you need.
theInputFiles = dir(filePattern);
filePattern = fullfile(outputFolder, '*.png'); % Change to whatever pattern you need.
theOutputFiles = dir(filePattern);

% Get the first file in the input folder.
baseInputFileName = theInputFiles(1).name;
fullInputFileName = fullfile(theInputFiles(1).folder, baseInputFileName);
inputImage = imread(fullInputFileName);
for k = 1 : length(theOutputFiles)
	baseOutputFileName = theOutputFiles(k).name;
	fullOutputFileName = fullfile(theInputFiles(k).folder, baseOutputFileName);
	fprintf(1, 'Now comparing %s to %s\n', fullInputFileName, fullOutputFileName);
	% Now do whatever you want with this file name,
	% such as reading it in as an image array with imread()
	imageArray = imread(fullOutputFileName);
	imshow(imageArray);  % Display image.
	drawnow; % Force display to update immediately.
	ssimResults(k) = ssim(imageArray, inputImage);
	immseResults(k) = immse(imageArray, inputImage);
% 	psnrResults(k) = psnr(imageArray, inputImage);
end