Raw sensor image and video de-hazing and atmospheric light analysis methods and systems
AssigneeThe Regents of the University of California
InventorsTruong Q. Nguyen, Yeejin Lee, Changyoung Han
Life of this patent
Estimating the term…
Abstract
A method for processing image or video data receives color filtered mosaiced raw image or video data. The method, performed in an image processing pipeline, de-hazes the color filtered mosaiced raw image or video data, e.g. Bayer image data, to create de-hazed mosaiced image or video data and de-mosaicing the de-hazed mosaiced image or video data to crate de-hazed and de-mosaiced image or video data.
Claims22
A method for processing image or video data performed in an image processing pipeline, the method comprising:
- receiving color filtered mosaiced raw image or video data, de-hazing the color filtered mosaiced raw image or video data to create de-hazed mosaiced image or video data and
- de-mosaicing the de-hazed mosaiced image or video data to create de-hazed and de-mosaiced image or video data, wherein the de-hazing comprises de-hazing a single data set of a single raw image,
- wherein the de-hazing comprises estimating atmospheric light in a subsampled version of the single raw image, and wherein said estimating applies a first assumption that atmospheric light varies smoothly in local neighborhood regions and the pixel values are similar to each other in the local neighborhood regions, and applies a second assumption that edges and details need not be preserved while estimating atmospheric light.
A method for processing image or video data performed in an image processing pipeline, the method comprising:
- receiving color filtered mosaiced raw image or video data, de-hazing the color filtered mosaiced raw image or video data to create de-hazed mosaiced image or video data and de-mosaicing the de-hazed mosaiced image or video data to create de-hazed and de-mosaiced image or video data, wherein the de-hazing comprises de-hazing a single data set of a single raw image, and
- wherein the de-hazing comprises estimating haze in a local neighborhood region of the single raw image and de-hazing the entire single raw image from estimated haze in the local neighborhood region.
A method for processing image or video data performed in an image processing pipeline, the method comprising:
- receiving color filtered mosaiced raw image or video data, de-hazing the color filtered mosaiced raw image or video data to create de-hazed mosaiced image or video data and
- de-mosaicing the de-hazed mosaiced image or video data to create de-hazed and de-mosaiced image or video data, wherein the de-hazing comprises de-hazing a single data set of a single raw image,
- wherein the de-hazing treats the color filtered mosaic data as piecewise constant and replaces missing pixels in the color filtered mosaic data with nearest neighbor pixels, and
- wherein:
- the de-hazing defines a new color channel region as a 2×2 block (connected 4-pixel to upper left location) of the raw image so that each of a R, G and B color value can be measured in the raw image;
- measures dark channels dpB the raw image within the same spatial region ΩpB of the raw image
- d
- p
- B
- =
- min
- p
- ∈
- Ω
- p
- B
-
- x
- p
- B
- x
- p
- B
- =
- ∑
- c
- ∈
- {
- r
- ,
- g
- ,
- b
- }
-
- x
- p
- s
-
- (
- c
- )
- where xpS is the raw image sampled by the sensor according to a pattern of the color filter;
- determines transmission tdB in the raw image as
- t
- d
- B
- =
- 1
- -
- γ
-
- min
- p
- ∈
- Ω
- p
- B
-
- y
- p
- B
- a
- B
- where ypB is divided by the corresponding atmospheric light color aB measured in pattern of the color filter;
- and
- estimates the c-th color of atmospheric light individually estimated from either of the most haze-opaque regions using bright channels or dark channels.
Dependencies are inferred from the claim text.