fitniqe

Fit custom model for NIQE image quality score

Syntax

model = fitniqe(imds)

model = fitniqe(imds,Name=Value)

Description

model = fitniqe(imds) creates a Naturalness Image Quality Evaluator (NIQE) model from reference image datastore imds.

model = fitniqe(imds,Name=Value) creates a NIQE model using one or more name-value arguments to control the model calculation.

Examples

Calculate NIQE Score Using Custom Feature Model

Load a set of natural images into an image datastore. These images are shipped in Image Processing Toolbox™ in a directory named 'imdata'.

setDir = fullfile(toolboxdir('images'),'imdata');
imds = imageDatastore(setDir,'FileExtensions',{'.jpg'});

Train a custom NIQE model using the image datastore.

model = fitniqe(imds);

Extracting features from 33 images.
.......
Done.

Read an image of a natural scene. Display the image.

I = imread('car1.jpg');
imshow(I)

Calculate the NIQE score for the image using the custom model. Display the score.

niqeI = niqe(I,model);
fprintf('NIQE score for the image is %0.4f.\n',niqeI)

NIQE score for the image is 2.6700.

Fit Custom NIQE Model Using Specified Block Size

Load a set of natural images into an image datastore. These images are shipped in Image Processing Toolbox™ in a directory named "imdata".

setDir = fullfile(toolboxdir("images"),"imdata");
imds = imageDatastore(setDir,FileExtensions=".jpg");

Create the custom model of NSS features using the image datastore. Specify a block size and use the default sharpness threshold.

model = fitniqe(imds,BlockSize=[48 96]);

Extracting features from 33 images.
.......
Done.

Read a natural image into the workspace. Display the image.

I = imread("yellowlily.jpg");
imshow(I)

Calculate the NIQE score for the image using the custom model. Display the score.

niqeI = niqe(I,model);
disp("NIQE score for the image is "+niqeI)

NIQE score for the image is 3.0313

Input Arguments

imds — Reference image datastore

ImageDatastore object

Reference image datastore, specified as an ImageDatastore object. The datastore must contain 2-D grayscale or 2-D RGB images of data type single, double, int16, uint8, or uint16.

Name-Value Arguments

Specify optional pairs of arguments as Name1=Value1,...,NameN=ValueN, where Name is the argument name and Value is the corresponding value. Name-value arguments must appear after other arguments, but the order of the pairs does not matter.

Example: model = fitniqe(imds,BlockSize=[48 36]) fits a NIQE model using 48-by-36 pixel blocks.

Before R2021a, use commas to separate each name and value, and enclose Name in quotes.

Example: model = fitniqe(imds,"BlockSize",[48 36]) fits a NIQE model using 48-by-36 pixel blocks.

BlockSize — Block size used to partition the images

[96 96] (default) | 2-element row vector of positive even integers

Block size used to partition the images, specified as a 2-element row vector of positive even integers. Blocks are nonoverlapping. Natural scene statistics, which are calculated from the blocks, define the output model.

Data Types: single | double | int8 | int16 | int32 | uint8 | uint16 | uint32

SharpnessThreshold — Sharpness threshold

0 (default) | numeric scalar in the range [0, 1]

Sharpness threshold, specified as a numeric scalar in the range [0, 1]. The sharpness threshold, s, controls which image blocks are used to compute the model. fitniqe computes the model using all blocks that have sharpness more than s times the maximum sharpness among all blocks.

Data Types: single | double

Output Arguments

model — Custom model of image features

niqeModel object

Custom model of image features, returned as a niqeModel object.

Tips

• The custom data set specified in the image datastore imds should consist of images that are perceptually pristine to human subjects. However, the definition of pristine depends on the application. For example, a pristine set of microscopy images has a different set of quality criteria than images of buildings or outdoor scenes. When training a custom NIQE model, use images with varied image content and with potentially different sets of quality criteria.

References

[1] Mittal, A., R. Soundararajan, and A. C. Bovik. "Making a Completely Blind Image Quality Analyzer." IEEE Signal Processing Letters. Vol. 22, Number 3, March 2013, pp. 209–212.

Version History

Introduced in R2017b