Hello, every one,
could you tell me how to measure the similarity degree of different pictures? For example, how much is the similarity degree of two dog pictures, or a simpler scene, how much is the similarity degree of a line and a curve in different pictures?
Dear Yiming,
First, you need to extract features from both images and then compare the two feature vectors using any similarity measure, for example, Euclidean distance.
This is a more complex question than you seem to think! On what basis do you want to measure similarity? Luminance? Frequency spectrum? Semantic content? The last one is hard. The others are a matter of selecting the tool you want to use.
The Structural SIMiliarity (SSIM) index is one option:
http://live.ece.utexas.edu/research/quality/SSIM/
And you can access the functionality in Matlab (among others) using the SHINE toolbox:
Willenbockel, et al., 2010. Controlling low-level image properties: The SHINE toolbox. Beh Res Methods, 42 (3)
You can do this at feature level. Initially extract salient shape, color or texture features and represent each image as a k-dimensional feature vector. Then by finding distance you can compute degree of similarity.
There are many methods to do this. The general idea is to extract features (i.e. parameters that describe the image as accurate as possible), and arrange it in a vector form. Then perform a comparison among these vectors.
The feature extraction is a task whose solution depends on the image to be analyzed.
CBIR (Content Based Image Retrieval) is a common technique used in computer vision field.
Nevertheless, you can use methods based on PCA (Principal Component Analysis), Wavelet Transformer, and others.
In most cases, the extracted feature vectors are used to train a machine learning-based classifiers, such as: neural network, decision tree, Naive Bayes, etc.
I applied these techniques to detect flaws in aerospace materials.
You can see my papers:
G.D’Angelo, S. Rampone, “Feature Extraction and Soft Computing Methods for Aerospace Structure Defect Classification,” Measurement, Vol. 85, pp.192-209, May 2016.
DOI:10.1016/j.measurement.2016.02.027.
G.D’Angelo, M. Laracca, S. Rampone, “Automated Eddy Current Non-Destructive Testing through Low Definition Lissajous Figures,” Proceedings of the 3rd IEEE International Workshop on Metrology for Aerospace, Florence, Italy, June 22-23, 2016.
DOI: 10.1109/MetroAeroSpace.2016.7573227
http://ieeexplore.ieee.org/document/7573227/?reload=true
Article Feature Extraction and Soft Computing Methods for Aerospace ...
fisrt, you should calculate a descriptor such as histogram, SIFT, SURF to descrbe each picture, then a vectors of numbers will generated.
the second step is to use any distance (eucliden, ...) to calculate difference between the both pictures
Hi,
SSIM will help you , check this
http://live.ece.utexas.edu/research/quality/SSIM/
similarity between two pictures is quantified in terms of a distance measure which is defined on the corresponding multi-dimensional feature space. Common distance measures are: the Minkowski distance, the Manhattan distance, the Euclidean distance and the Hausdorff distance.
https://link.springer.com/chapter/10.1007%2F978-3-642-11216-4_14
https://datascience.stackexchange.com/questions/48642/how-to-measure-the-similarity-between-two-images
ssimval = ssim(A,ref) computes the Structural Similarity Index (SSIM) value for image A using ref as the reference image.
peaksnr = psnr(A,ref) calculates the peak signal-to-noise ratio for the image A, with the image ref as the reference. A and ref must be of the same size and class.
err = immse(X,Y) calculates the mean-squared error (MSE) between the arrays X and Y. X and Y can be arrays of any dimension, but must be of the same size and class.
Dear;
for more information please check the following documents:
https://in.mathworks.com/help/images/ref/ssim.html
Regards
Dear;
for more information please check the following documents:
http://live.ece.utexas.edu/research/quality/SSIM/
Regards
- Badges
- Science topic
- Similar topics
- Control Systems Engineering
- Filtering