Image quality assessment using similar scenes as reference

1. A method for image quality assessment of non-aligned images, comprising the steps of

providing non-aligned reference and distorted images;

inputting reference images to a first deep path of a convolutional neural network having a set of parameters;

inputting distorted images to a second deep path of the convolutional neural network sharing a set of parameters with the first deep path convolutional neural network;

sharing weights between the first and second deep path convolutional neural networks to support extraction of a same set of features;

in a fully connected layer connected to receive input from both the first and second deep paths of the convolutional neural network, generating an image quality assessment as a linear regressor; and

outputting an image quality score.

2. The method for image quality assessment of non-aligned images of claim 1 , wherein both the first and second deep paths of the convolutional neural network have multiple respective convolution layers, and at least one respective pooling layer.

3. The method for image quality assessment of non-aligned images of claim 1 , wherein both the first and second deep paths of the convolutional neural network feed into a concatenation layer to concatenate features from the first and second deep paths of the convolutional neural networks.

4. The method for image quality assessment of non-aligned images of claim 1 , wherein both the first and second deep paths of the convolutional neural network feed into a concatenation layer to concatenate features from the first and second deep paths of the convolutional neural network and discriminative features are mapped in fully connected layers to generate the image quality assessment as a linear regressor.

5. The method for image quality assessment of non-aligned images of claim 1 , wherein the image quality score is output by minimizing a Euclidean loss.

6. A system for image quality assessment of non-aligned images, comprising

a first deep path portion of a convolutional neural network having a set of parameters;

a second deep path portion of the convolutional neural network sharing a set of parameters with the first deep path convolutional neural network, wherein weights are shared between the first and second deep paths of the convolutional neural network to support extraction of a same set of features, and wherein non-aligned reference and distorted images are respectively provided to the first and second deep paths of the convolutional neural network for processing;

a concatenation layer connected to both the first and second deep paths convolutional neural network;

a fully connected layer connected to the concatenation layer to receive input from both the first and second deep paths of the convolutional neural network, generating an image quality assessment as a linear regressor; and

outputting an image quality score.

7. The system for image quality assessment of non-aligned images of claim 6 , wherein both the first and second deep paths of the convolutional neural network have multiple respective convolution layers, and at least one respective pooling layer.

8. The system for image quality assessment of non-aligned images of claim 6 , wherein the concatenation layer concatenates features from the first and second deep paths of the convolutional neural network and discriminative features are mapped in fully connected layers to generate the image quality assessment as a linear regressor.

9. The system for image quality assessment of non-aligned images of claim 6 , wherein the image quality score is output by minimizing a Euclidean loss.

10. A method for generating a training set for image quality assessment of non-aligned images, comprising the steps of using raw images and providing non-aligned reference and distorted images by synthesizing a set of non-aligned images;

inputting reference images to a first deep path of a convolutional neural network having a set of parameters;

inputting distorted images to a second deep path of the convolutional neural network sharing a set of parameters with the first deep path convolutional neural network;

sharing weights between the first and second deep path of the convolutional neural network to support extraction of a same set of features in each neural network;

in a fully connected layer connected to receive input from both the first and second deep paths of the convolutional neural network, generating an image quality assessment as a linear regressor; and

outputting an image quality score.

11. The method for image quality assessment of non-aligned images of claim 10 , wherein both the first and second deep paths of the convolutional neural network have multiple respective convolution layers, and at least one respective pooling layer.

12. The method for image quality assessment of non-aligned images of claim 10 , wherein both the first and second deep paths of the convolutional neural network feed into a concatenation layer to concatenate features from the first and second deep paths of the convolutional neural network.

13. The method for image quality assessment of non-aligned images of claim 10 , wherein both the first and second deep paths of the convolutional neural network feed into a concatenation layer to concatenate features from the first and second deep paths of the convolutional neural network and discriminative features are mapped in fully connected layers to generate the image quality assessment as a linear regressor.

14. The method for image quality assessment of non-aligned images of claim 10 , wherein the image quality score is output by minimizing a Euclidean loss.