High-dynamic-range image generation with pre-combination denoising

1. A method for generating a high-dynamic-range (HDR) image, comprising:

denoising a short-exposure-time image, wherein denoising comprises

applying a first guided filter to the short-exposure-time image, the guided filter utilizing a long-exposure-time image as its guide,

after the step of denoising, applying a second guided filter to the long-exposure-time image, the second guided filter utilizing the short-exposure-time image as its guide, and scaling at least one of the short-exposure-time image and the long-exposure-time image to place the short-exposure-time image and the long-exposure-time image on a common radiance scale; and

after the step of scaling, merging the short-exposure-time image with the long-exposure-time image to generate the HDR image.

2. The method of claim 1 , the step of denoising comprising applying a position-sensitive filter to the short-exposure-time image such that, for each local region of the short-exposure-time image, the position-sensitive filter is dependent on pixel values of a corresponding local region of the long-exposure-time image.

3. The method of claim 1 , wherein the step of applying the first guided filter comprises determining a pixel value in the denoised short-exposure-time image as a linear function of a corresponding pixel value in the long-exposure-time image in a local region of the short-exposure-time image at said pixel value in the denoised short-exposure-time image.

4. The method of claim 1 , wherein the step of applying the first guided filter comprises determining a i′-th pixel value in the denoised short-exposure-time image S as S′ S,i =aS L,i +b, wherein S L,i is the pixel value of the long-exposure-time image, and wherein each of coefficients a and b depends on pixel values in a local region of the short-exposure-time image at the i′-th pixel.

5. The method of claim 4 , wherein each of the coefficients a and b are position sensitive.

6. The method of claim 1 , the first guided filter being selected from the group of filters comprising: a joint bilateral filter and a joint guided filter.

7. The method of claim 1 , wherein the step of applying the second guided filter comprises determining a pixel value in the denoised long-exposure-time image as a linear function of a corresponding pixel value in the short-exposure-time image in a local region of the long-exposure-time image at said pixel value in the denoised long-exposure-time image.

8. The method of claim 1 , the step of applying the second guided filter comprising determining an i′-th pixel value in the denoised long-exposure-time image L as S′ L,i =c·S S,i +d, wherein S S,i is the pixel value of the short-exposure-time image and each of coefficients c and d depends on pixel values in a local region of the long-exposure-time image at the i′-th pixel.

9. The method of claim 8 , wherein the coefficients c and d are position sensitive.

10. The method of claim 1 , the second guided filter being selected from the group of filters comprising: a joint bilateral filter and a joint guided filter.

11. The method of claim 1 , further comprising adjusting a spatial scale of the second guided filter to adjust a degree to which noise properties of the short-exposure-time image are blended into the long-exposure-time image.

12. The method of claim 1 , further comprising:

evaluating a degree of discrepancy between the short-exposure-time image and the long-exposure time image; and

in the step of merging, generating the HDR image as a linear combination of the short-exposure-time image and the long-exposure-time image using combination weights depending at least in part on the degree of discrepancy.

13. The method of claim 1 , further comprising obtaining the short-exposure-time image and the long-exposure-time image from a single frame captured by a spatially-multiplexed-exposure image sensor.

14. The method of claim 1 , each of the short-exposure-time image and the long-exposure-time image being a color image composed of a plurality of color components, the method comprising:

performing the steps of denoising, scaling, and merging separately for each of the plurality of color components to generate a plurality of color-specific HDR images for the plurality of color components, respectively; and

demosaicing the plurality of color-specific HDR images to create a full-color HDR image.

15. The method of claim 1 , each of the short-exposure-time image and the long-exposure-time image being a color image composed of a plurality of color components, the color components including a first color, the method comprising:

for each of the color components, performing the step of denoising, based at least partly on the first color of the long-exposure-time image;

after performing the step of denoising for each of the color components, performing the steps of scaling and merging separately for each of the color components separately to generate a plurality of color-specific HDR images for the plurality of color components, respectively; and

demosaicing the plurality of color-specific HDR images to create a full-color HDR image.

16. The method of claim 1 , further comprising:

receiving the short-exposure-time image and the long-exposure-time image from an image sensor; and

outputting the HDR image to a component selected from the group consisting of a display, a non-transitory memory, and a communication bus.

17. An HDR camera, comprising:

a spatially-multiplexed-exposure image sensor configured to capture, in a single frame, a short-exposure-time image and a long-exposure-time image;

a processor for executing the method of claim 1 .

18. A non-transitory computer-readable storage medium having stored thereon computer-executable instructions for executing with one or more processors a method in accordance the method of claim 1 .