Sparse Feature Fidelity (SFF) for Image Quality Assessment

A Matlab code of the Sparse Feature Fidelity (SFF) for perceptual image quality assessment
882 Downloads
Updated 31 Dec 2013

View License

SFF is a new algorithm for evaluating perceptual quality of color images.

For quality evaluation, you can just run 'SFF' as follows:

load('W.mat'); % load the feature detector
score = SFF(refImg, disImg, W); % refImg and disImg respectively denote the reference image and distorted image

The quality scores are between 0 and 1, where 1 represents the same quality as the reference image.

Feature detector W is a matrix of size 8*192 generated by running TrainW(18000,8,8) on data1. W can be used for extracting features from image patches. The Training code is in the folder '\Training'. You can run 'TrainW' to get the feature detector. For example:

W = TrainW(18000,8,8);

In our paper we used 18000 sample patches of size 8*8, and retained only 8 components. Two sets of images are provided for the training stage, i.e., data1 and data2, which is described in our paper. We suggest you use data1 for training.

SFF performs very well on CSIQ, LIVE, IVC, TID2008, TID2013, and Toyama database. This package also provides four examples for testing this algorithm on CSIQ, LIVE, TID2008, and TID2013 databases.

Thorough discussion can be found in:
Hua-wen Chang, Hua Yang, Yong Gan, and Ming-hui Wang, "Sparse Feature Fidelity for Perceptual Image Quality Assessment", IEEE Transactions on Image Processing, vol. 22, no. 10, pp. 4007-4018, October 2013

Cite As

Hua-wen Chang (2024). Sparse Feature Fidelity (SFF) for Image Quality Assessment (https://www.mathworks.com/matlabcentral/fileexchange/43805-sparse-feature-fidelity-sff-for-image-quality-assessment), MATLAB Central File Exchange. Retrieved .

MATLAB Release Compatibility
Created with R2009b
Compatible with any release
Platform Compatibility
Windows macOS Linux

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!
Version Published Release Notes
1.3.0.0

Add new results of TID2013 and testing codes on TID2008 and TID2013

1.2.0.0

Experimental results on five databases are added.

1.0.0.0