Neighborhood haze density estimation for single-image dehaze

Patent metadata

Field	Value
Publication number	US-10477128-B2
Application number	US-201815865058-A
Country	US
Kind code	B2
Filing date	Jan 8, 2018
Priority date	Jan 6, 2017
Publication date	Nov 12, 2019
Grant date	Nov 12, 2019

Abstract

Dehazed images are produced based on an atmospheric light image obtained form an input image as brightest pixels of a predetermined window and white map. The white map is median filtered, morphologically filtered and, in some examples, filtered with a guided filter, and the filtered image combined with the atmospheric light image to produce a dehazed image.

First claim

We claim:

1. An image processor, comprising a memory storing processor-executable instructions for a method comprising:
processing an input image to produce a bright channel image associated with largest brightness values in the input image over a first selected window;
defining an atmospheric light image corresponding to atmospheric light based on the bright channel image;
processing the input image to produce a white map image based on input image brightness and saturation values;
producing a transmission map image based on the white map image, the atmospheric light image, and a haze value h, wherein 0≤h≤1; and
determining a dehazed image based on the input image, the transmission map image, and the atmospheric light image.

2. The image processor of claim 1 , wherein the method further comprises obtaining the atmospheric light image by applying a guided filter to the bright channel image.

3. The image processor of claim 2 , wherein the guided filter is based on the bright channel image as a guidance image.

4. The image processor of claim 2 , wherein the white map image based on the input image corresponds to (1−S)/L for each pixel, wherein L is a brightness value and S is a saturation value associated with the pixel.

5. The image processor of claim 1 , wherein the transmission map image is based on a median filtered, morphologically reconstructed white map image.

6. The image processor of claim 1 , wherein the transmission map image is based on a median filtered, morphologically reconstructed, and guided filtered white map image.

7. The image processor of claim 1 , wherein the dehazed image is produced as (J−A)/T+A, wherein J is the input image, A is the atmospheric light image, and T is the transmission image.

8. The image processor of claim 1 , wherein the first selected window is a rectangular area between 2 and 200 pixels wide and 2 and 200 pixels high.

9. The image processor of claim 5 , wherein the median filter is based on a second selected window having a rectangular area between 2 and 200 pixels wide and 2 and 200 pixels high.

10. The image processor of claim 9 , wherein the first window is 10 pixels wide.

11. The image processor of claim 1 , further comprising processing the input image to obtain image brightness and saturation values from the input image.

12. The image processor of claim 1 , further comprising displaying the dehazed image.

13. An image processing method, comprising:
with a processor,
producing a smoothed image and an atmospheric light image based on an input image wherein the atmospheric light image is produced by processing the input image to produce a bright channel image associated with largest brightness values in the input image in associated windows, each of the associated windows having a common size; and
dehazing the input image by combining the smoothed image, the atmospheric light image, and the input image.

14. The image processing method of claim 13 , further comprising median filtering the input image, wherein the smoothed image is produced by morphologically reconstructing the median filtered input image.

15. An image processing method, comprising:
with a processor,
processing the input image to produce a bright channel image;
producing a smoothed image and an atmospheric light image based on the input image; and
dehazing the input image by combining the smoothed image, the atmospheric light image, and the input image, wherein the atmospheric light image is produced by applying a guided filter to the bright channel image.

16. The image processing method of claim 15 , further comprising:
producing a basemap image B=1−W/A, wherein W is the smoothed image and A is the atmospheric light image; and
defining a transmission map image T=hB+(1−h), wherein h is a haze factor 0≤h≤1, and the input image is dehazed to produce a dehazed output image as (J−A)/T+A, wherein J is the input image.

17. The image processing method of claim 15 , further wherein the input image is dehazed to produce a dehazed output image as (J−A)/T+A, wherein T=hB+(1−h), h is a haze factor 0≤h≤1, B=1−W/A, W is the smoothed image, J is the input image, and A is the atmospheric light image.

18. The image processing method of claim 17 , wherein the smoothed image W is produced by applying a median filter and morphological reconstruction to the input image.

19. The image processing method of claim 18 , wherein the atmospheric light image is produced by processing the input image to produce a bright channel image associated with largest brightness values in the input image over a selected window.

20. The image processing method of claim 18 , wherein the input image, the bright channel image, the atmospheric light image, the smoothed image, the basemap image, and the transmission map image are three channel images, wherein each channel of the bright channel image is associated with a largest brightness value in the input image in the corresponding channel.

21. The image processing method of claim 20 , wherein the input image, the bright channel image, the atmospheric light image, the smoothed image, the basemap image, and the transmission map image are RGB images.

22. The method of claim 15 , wherein the smoothed image is produced by sequentially applying a median filter, a morphological filter, and a guided filter to the input image.

23. The image processing method of claim 15 , further comprising median filtering the input image, wherein the smoothed image is produced by morphologically reconstructing the median filtered input image.

24. An image acquisition system, comprising:
an image sensor;
memory coupled to the image sensor so as to store an image; and
a processor coupled to the memory so as to retrieve the image and produce a dehazed image based on pixel by pixel processing that produces a first image corresponding to morphologically filtering the input image and a second image corresponding to brightest pixels in the image over a predetermined window.

25. The image acquisition system of claim 24 , wherein the processor is coupled to produce an atmospheric light image based on the second image, and produce the dehazed image based on the morphologically filtered input image and the atmospheric light image.

26. The image acquisition system of claim 25 , wherein the atmospheric light image is produced by applying a guided filter to second image.

Assignees

• Nikon Corp

Inventors

• Bhutani Ripul
• Chang Ping-Wei
• Yuan Bausan

Classifications

• H04N9/646

  for image enhancement, e.g. vertical detail restoration, cross-colour elimination, contour correction, chrominance trapping filters · CPC title

• H04N1/6027 Primary

  Correction or control of colour gradation or colour contrast (H04N1/6058 takes precedence) · CPC title

• H04N25/674 Primary

  based on the scene itself, e.g. defocusing · CPC title

• H04N23/63

  by using electronic viewfinders · CPC title

• H04N23/81

  for suppressing or minimising disturbance in the image signal generation · CPC title