System and method for learning-based image super-resolution

1. An image processing system, comprising:

an input interface to receive a set of pairs of images to train an image generator, each pair includes a low-resolution image of a scene and a high-resolution image of the scene;

a processor to train the image generator by solving an optimization problem to produce parameters of the image generator reducing distances between image gradients of the high-resolution images and the corresponding low-resolution images upsampled by the image generator; and

an output interface to render the parameters of the image generator.

2. The image processing system of claim 1 , wherein the image generator is a neural network, wherein the parameters of the image generator are weights of connections between nodes of different layers of the neural network, such that the processor trains the neural network using the set of pairs of images.

3. The image processing system of claim 1 , wherein the processor minimizes a cost function including a weighted combination of the distances between the image gradients and distances between pixel intensities of the high-resolution images and the corresponding low-resolution images upsampled by the image generator.

4. The image processing system of claim 1 , wherein the processor solves the optimization problem using a stochastic gradient descent.

5. The image processing system of claim 1 , wherein the processor is configured to compute the image gradients using a Gaussian kernel convolution followed by spatial gradient calculation.

6. The image processing system of claim 1 , wherein the processor determines the image gradients on a pixel-by-pixel basis.

7. A device operatively connected to the image generator trained by the image processing system of claim 1 , wherein the device is configured to upsample an input image using the image generator and perform a control action based on the upsampled input image.

8. An image processing method, wherein the method uses a processor coupled with stored instructions implementing the method, wherein the instructions, when executed by the processor carry out steps of the method, comprising:

receiving a set of pairs of images to train an image generator, each pair includes a low-resolution image of a scene and a high-resolution image of the scene;

training the image generator by solving an optimization problem to produce parameters of the image generator reducing distances between image gradients of the high-resolution images and the corresponding low-resolution images upsampled by the image generator; and

outputting the parameters of the image generator.

9. The image processing method of claim 8 , wherein the image generator is a neural network.

10. The image processing method of claim 8 , wherein the solving comprises:

minimizing a cost function including a weighted combination of the distances between the image gradients and distances between pixel intensities of the high-resolution images and the corresponding low-resolution images upsampled by the image generator.

11. The image processing method of claim 8 , wherein the optimization problem is solved using a stochastic gradient descent.

12. The image processing method of claim 8 , further comprising:

computing the image gradients using a Gaussian kernel convolution followed by spatial gradient calculation.

13. The image processing method of claim 8 , wherein the image gradients are computed on a pixel-by-pixel basis.

14. A non-transitory computer readable storage medium embodied thereon a program executable by a processor for performing a method, the method comprising:

receiving a set of pairs of images to train an image generator, each pair includes a low-resolution image of a scene and a high-resolution image of the scene;

training the image generator by solving an optimization problem to produce parameters of the image generator reducing distances between image gradients of the high-resolution images and the corresponding low-resolution images upsampled by the image generator; and

outputting the parameters of the image generator.

15. The medium of claim 14 , wherein the solving comprises:

minimizing a cost function including a weighted combination of the distances between the image gradients and distances between pixel intensities of the high-resolution images and the corresponding low-resolution images upsampled by the image generator.