Image processing method and device

1. An image processing method, comprising:

obtaining a first image;

obtaining a second image;

extracting a first vertical gradient and a first horizontal gradient of each band in the first image using a first gradient operator;

extracting a second vertical gradient and a second horizontal gradient of each band in the second image using a second gradient operator;

calculating a first gradient saliency feature map of each band in the first image based on the first vertical gradient and the first horizontal gradient;

performing weighted averaging on each pixel in the first gradient saliency feature map of each band in the first image to obtain a first gradient saliency map of the first image;

calculating a second gradient saliency feature map of each band in the second image based on the second vertical gradient and the second horizontal gradient;

performing weighted averaging on each pixel in the second gradient saliency feature map of each band in the second image to obtain a second gradient saliency map of the second image;

performing, based on the first gradient saliency feature map and the second gradient saliency feature map, feature matching between the first image and the second image to obtain a dense disparity map of the first image;

performing disparity translation on a second pixel in the second image corresponding to a first pixel in the first image based on the dense disparity map; and

fusing, based on the dense disparity map, the first image and the second pixel.

2. The image processing method of claim 1 , wherein after obtaining the first image and obtaining the second image, the image processing method further comprises adjusting the first image or the second image such that there is only a horizontal disparity between the first image and the second image.

3. The image processing method of claim 1 , further comprising searching, in the second image along a disparity direction, for a matching unit in the first image to obtain a disparity value that has a minimum feature difference.

4. The image processing method of claim 3 , wherein the matching unit is a neighborhood of each pixel in the first image, and wherein the image processing method further comprises:

using the disparity value for the first pixel; and

performing feature matching on all pixels in the first image to obtain the dense disparity map of the first image.

5. The image processing method of claim 3 , wherein the first image comprises a plurality of superpixels, wherein the matching unit is one of the superpixels, and wherein the image processing method further comprises searching for each of the superpixels in the second image along the disparity direction to obtain the disparity value.

6. The image processing method of claim 1 , wherein the first image is a near infrared image, wherein the second image is a color visible light image, and wherein the image processing method further comprises:

translating first color information of a pixel that corresponds to each pixel in the near infrared image and that is in the color image to obtain second color information of a fused image based on a corresponding disparity value, based on the dense disparity map, and for each pixel in the near infrared image; and

performing weighted averaging on a first luminance value of a pixel of the color image and a second luminance value of the near infrared image to obtain luminance information of the fused image based on the dense disparity map and for each pixel in the near infrared image.

7. The image processing method of claim 1 , wherein the first image is a color visible light image, wherein the second image is a thermal infrared image, and wherein the image processing method further comprises:

translating luminance information of a pixel that corresponds to each pixel in the color image and that is in the thermal infrared image based on the dense disparity map, based on an estimated disparity, and for each pixel in the color image; and

performing weighting on a first luminance value of the thermal infrared image and a second luminance value of the color image to obtain a third luminance value of a fused image or normalizing the first luminance value and multiplying the first luminance value by the second luminance value at a corresponding location to obtain the third luminance value.

8. The image processing method of claim 1 , wherein the first image is a color visible light image, wherein the second image is a dynamic vision sensor (DVS) texture image, and wherein the image processing method further comprises:

translating luminance information of a pixel that corresponds to each pixel in the color image and that is in the DVS texture image based on the dense disparity map, based on an estimated disparity, and for each pixel in the color image;

performing wavelet decomposition or pyramid decomposition on a first luminance component of the color image to divide the first luminance component into a low-frequency component and a high-frequency component;

performing weighting on the high-frequency component and the DVS texture image to obtain a new high-frequency component; and

performing an inverse transform operation on the low-frequency component to obtain a second luminance component of a fused image.

9. The image processing method of claim 1 , wherein the first image is a multispectral image (MSI), wherein the second image is a hyperspectral image (HSI), and wherein the image processing method further comprises:

translating a pixel that corresponds to each pixel in the MSI image and that is in the HSI image based on the dense disparity map, based on an estimated disparity, and for each pixel in the MSI image;

performing radiation calibration on the MSI image;

performing radiation calibration on the HSI image;

extracting a first spectral endmember of the MSI and a second spectral endmember of the HSI using an endmember extraction method;

calculating a first initial abundance matrix of the MSI;

calculating a second initial abundance matrix of the HSI;

performing combined unmixing using a space spectrum correlation of the MSI and the HSI;

updating the first spectral endmember, the second spectral endmember, and the abundance matrices to convergence using multiplication iterations; and

multiplying an obtained high spatial resolution abundance matrix by a hyper-spectral endmember to obtain a fused image.

10. The image processing method of claim 1 , wherein the first image comprises a plurality of superpixels, wherein a matching unit along a disparity direction in the second image is one of the plurality of superpixels, and wherein the image processing method further comprises searching for each of the plurality of superpixels in the second image along the disparity direction to obtain the disparity value.

11. An image processing apparatus, comprising:

a sensor configured to:

generate a first image; and

generate a second image;

a chip coupled to the sensor and configured to:

receive the first image from the sensor;

receive the second image from the sensor;

extract a first vertical gradient and a first horizontal gradient of each band in the first image using a first gradient operator;

extract a second vertical gradient and a second horizontal gradient of each band in the second image using a second gradient operator;

calculate a first gradient saliency feature map of each band in the first image based on the first vertical gradient and the first horizontal gradient;

perform weighted averaging on each pixel in the first gradient saliency feature map of each band in the first image to obtain a first gradient saliency map of the first image;

calculate a second gradient saliency feature map of each band in the second image based on the second vertical gradient and the second horizontal gradient;

perform weighted averaging on each pixel in the second gradient saliency feature map of each band in the second image to obtain a second gradient saliency map of the second image;

perform, based on the first gradient saliency feature map and the second gradient saliency feature map, feature matching between the first image and the second image to obtain a dense disparity map of the first image;

perform disparity translation on a pixel in the second image corresponding to a pixel in the first image based on the dense disparity map of the first image; and

fuse the first image and the pixel in the second image.

12. The image processing apparatus of claim 11 , wherein after the chip receives the first image and the second image, and the chip is further configured to adjust the first image or the second image such that there is a horizontal disparity between the first image and the second image.

13. The image processing apparatus of claim 11 , wherein the chip is further configured to search, in the second image along a disparity direction, for a matching unit in the first image, to obtain a disparity value that has a minimum feature difference.

14. The image processing apparatus of claim 13 , wherein the matching unit is a neighborhood of each pixel in the first image, and wherein the chip is further configured to:

use the disparity value for the first pixel; and

perform feature matching on all pixels in the first image to obtain the dense disparity map of the first image.

15. The image processing apparatus of claim 13 , wherein the first image comprises a plurality of superpixels, wherein the matching unit is one of the superpixels, and wherein the chip is further configured to search for each of the superpixels in the second image along the disparity direction to obtain the disparity value.

16. The image processing apparatus of claim 11 , wherein the first image is a near infrared image, wherein the second image is a color visible light image, and wherein the chip is further configured to:

translate first color information of a pixel that corresponds to each pixel in the near infrared image and that is in the color image to obtain second color information of a fused image based on the dense disparity map and for each pixel in the near infrared image; and

perform weighted averaging on a first luminance value of a pixel of the color image and a second luminance value of the near infrared image to obtain luminance information of the fused image based on the corresponding disparity value, the dense disparity map, and for each pixel in the near infrared image.

17. The image processing apparatus of claim 11 , wherein the first image is a color visible light image, wherein the second image is a thermal infrared image, and wherein the chip is further configured to:

translate luminance information of a pixel that corresponds to each pixel in the color image and that is in the thermal infrared image based on the dense disparity map, an estimated disparity, and for each pixel in the color image; and

perform weighting on a first luminance value of the thermal infrared image and a second luminance value of the color image to obtain a third luminance value of a fused image or normalize the first luminance value and multiply the first luminance value by the second luminance value at a corresponding location to obtain the third luminance value.

18. The image processing apparatus of claim 11 , wherein the first image is a color visible light image, wherein the second image is a dynamic vision sensor (DVS) texture image, and wherein the chip is further configured to:

translate luminance information of a pixel that corresponds to each pixel in the color image and that is in the DVS texture image based on the dense disparity map, based on an estimated disparity, and for each pixel in the color image;

perform wavelet decomposition or pyramid decomposition on a first luminance component of the color image, to divide the first luminance component into a low-frequency component and a high-frequency component;

perform weighting on the high-frequency component and the DVS texture image to obtain a new high-frequency component; and

perform an inverse transform operation on the low-frequency component to obtain a second luminance component of a fused image.

19. The image processing apparatus of claim 11 , wherein the first image is a multispectral image (MSI), wherein the second image is a hyperspectral image (HSI), and wherein the chip is further configured to:

translate a pixel that corresponds to each pixel in the MSI image and that is in the HSI image based on the dense disparity map, based on an estimated disparity, and for each pixel in the MSI image;

perform radiation calibration on the MSI image;

perform radiation calibration on the HSI image;

extract a first spectral endmember of the MSI and a second spectral endmember of the HSI using an endmember extraction method;

calculate a first initial abundance matrix of the MSI;

calculate a second initial abundance matrix of the HSI;

perform combined unmixing using a space spectrum correlation of the MSI and the HSI;

update the first spectral endmember, the second spectral endmember, and the abundance matrices to convergence using multiplication iterations; and

multiply an obtained high spatial resolution abundance matrix by a hyper-spectral endmember to obtain a fused image.

20. The image processing apparatus of claim 11 , wherein the first image comprises a plurality of superpixels, wherein a matching unit along a disparity direction in the second image is one of the plurality of superpixels, and wherein the chip is further configured to search for each of the plurality of superpixels in the second image along the disparity direction to obtain the disparity value.