System and method for assessing the quality of a high-dynamic range (HDR) image

1 . A system for assessing quality of a high-dynamic range (HDR) image, comprising:

a feature extraction module arranged to extract a plurality of frequency features on a pair of reference image and a distorted image generated based on the reference image, wherein the plurality of frequency features includes:

a local frequency feature extracted using an odd log-Gabor filter applied to a luminance component of the image; and

a global frequency feature extracted from a frequency spectrum of the image using a Butterworth filter;

a comparison module arranged to compare a pair of feature maps obtained by processing the extracted frequency features on both the reference image and the distorted image, by:

generating a local similarity map by comparing the local frequency feature from the reference image and the distorted image; and

generating a global similarity map by comparing the global frequency feature from the reference image and the distorted image; and

a scoring module arrange to output an image quality assessment (IQA) score of the distorted image with reference to the reference image provided, by performing feature pooling to generate a local frequency similarity score and a global frequency similarity score associated with a weighted similarity of the distorted image comparing to the reference image based on the local similarity map and the global similarity map, and the IQA score is represented by a combination of the local frequency similarity score and the global frequency similarity score;

wherein the plurality of frequency features are associated with sensitive information in a human visual system (HVS).

2 . The system of claim 1 , wherein the plurality of frequency features are associated with structural information and/or partial frequencies when a visual scene presented by the image is perceived by a human observer.

3 . The system of claim 2 , wherein the local frequency feature represents texture details of an image perceived by human eyes.

4 . The system of claim 3 , wherein the odd log Gabor Filter is arranged to extract the local frequency feature associated with a high-frequency component of the image.

5 . The system of claim 4 , wherein the Gabor filter is arranged to extract horizontal and vertical edge features on the image, and wherein the local frequency feature is extracted by applying a spatial mask on the extracted horizontal and vertical edge features with high luminance.

6 . The system of claim 2 , wherein the global frequency feature characterizes a sensitive frequency interval of the HVS.

7 . The system of claim 6 , wherein the Butterworth filter is arranged to compose a frequency map and a phase map associated with image, wherein the frequency map and the phase map are obtained by:

obtaining a frequency spectrum representation by performing a Discrete Fourier Transform (DFT) to the image associated with a frequency representation of each pixel of the image in the spatial domain;

applying a bandpass Butterworth filter to provide high weights to a predetermined frequency interval; and

separating the frequency spectrum representation into real part and imaginary part to obtain the frequency map and the phase map.

8 . The system of claim 3 , further comprising an image pre-processing module arranged to transfer the reference image and the distorted image to a perceptual space to map a wide range of luminance to a perceptual range of the HVS.

9 . The system of claim 8 , wherein a luminance map for each of the reference image I r (x,y) and distorted image I d (x,y) are obtained from a linear luminance space, wherein (x,y) denotes the pixel coordinate in the image.

10 . The system of claim 9 , wherein the IQA score is represented as Q LGFM =Q G (x,y)·Q L (x,y), wherein Q G (x,y) and Q L (x,y) denotes the local frequency similarity score and global frequency similarity score calculated as an weighted average over all the pixel coordinates (x,y) on the corresponding similarity maps.

11 . A method for assessing quality of a high-dynamic range (HDR) image, comprising the steps of:

extracting a plurality of frequency features on a pair of reference image and a distorted image generated based on the reference image, wherein the plurality of frequency features includes:

a local frequency feature extracted using an odd log-Gabor filter applied to a luminance component of the image; and

a global frequency feature extracted from a frequency spectrum of the image using a Butterworth filter;

comparing a pair of feature maps obtained by processing the extracted frequency features on both the reference image and the distorted image, by:

generating a local similarity map by comparing the local frequency feature from the reference image and the distorted image; and

generate a global similarity map by comparing the global frequency feature from the reference image and the distorted image; and

outputting an image quality assessment (IQA) score of the distorted image with reference to the reference image provided, by performing feature pooling to generate a local frequency similarity score and a global frequency similarity score associated with a weighted similarity of the distorted image comparing to the reference image based on the local similarity map and the global similarity map, and the IQA score is represented by a combination of the local frequency similarity score and the global frequency similarity score;

wherein the plurality of frequency features are associated with sensitive information in a human visual system (HVS).

12 . The method of claim 11 , wherein the plurality of frequency features are associated with structural information and/or partial frequencies when a visual scene presented by the image is perceived by a human observer.

13 . The method of claim 11 , wherein the local frequency feature represents texture details of an image perceived by human eyes.

14 . The method of claim 13 , wherein the odd log-Gabor Filter is arranged to extract the local frequency feature associated with a high-frequency component of the image.

15 . The method of claim 14 , wherein the Gabor filter is arranged to extract horizontal and vertical edge features on the image, and wherein the local frequency feature is extracted by applying a spatial mask on the extracted horizontal and vertical edge features with high luminance.

16 . The method of claim 12 , wherein the global frequency feature characterizes a sensitive frequency interval of the HVS.

17 . The method of claim 16 , wherein the Butterworth filter is arranged to compose a frequency map and a phase map associated with image, wherein the frequency map and the phase map are obtained by:

obtaining a frequency spectrum representation by performing a Discrete Fourier Transform (DFT) to the image associated with a frequency representation of each pixel of the image in the spatial domain;

applying a bandpass Butterworth filter to provide high weights to a predetermined frequency interval; and

separating the frequency spectrum representation into real part and imaginary part to obtain the frequency map and the phase map.

18 . The method of claim 13 , further comprising a step of preforming pre-processing of the reference image and the distorted image by transferring the reference image and the distorted image to a perceptual space to map a wide range of luminance to a perceptual range of the HVS.

19 . The method of claim 18 , wherein the step of preforming pre-processing of the reference image and the distorted image further comprising the step of obtaining a luminance map for each of the reference image I r (x,y) and distorted image I d (x,y) from a linear luminance space, wherein (x,y) denotes the pixel coordinate in the image.

20 . The method of claim 19 , wherein the IQA score is represented as Q LGFM =Q G (x,y)·Q L (x,y), wherein Q G (x,y) and Q L (x,y) denotes the local frequency similarity score and global frequency similarity score calculated as an weighted average over all the pixel coordinates (x,y) on the corresponding similarity maps.