Method for accelerating Monte Carlo renders
AssigneeE-On Software
InventorsJulien Chauvier, Mauricio Delbracio, Nicholas Phelps
Life of this patent
Estimating the term…
Abstract
An image rendering method comprising the steps of: (a) performing a noisy estimation of a given view of a given scene of said image using a Monte-Carlo stochastic renderer, wherein general information of each ray sample, such as color and position in the image plane, is stored, (b) calculating a similarity measure between two sets of samples cast from two pixels of said scene and their neighbors, and (c) generating an output image wherein each pixel color is a weighted combination of the pixel colors generated from (a) with weights based on the similarity measure of (b) between the respective set of samples cast from the pixels and their neighbors.
Claims14
An image rendering method comprising the steps of:
- (a) performing, using a processor, a noisy estimation of a given view of a given scene of an image using a Monte-Carlo stochastic renderer, wherein general information of each ray sample is stored,
- (b) calculating a similarity measure between two sets of samples cast from two pixels of said scene and their neighbors, and
- (c) generating an output image wherein each pixel color is a weighted combination of the pixel colors generated from the step (a) with weights based on the similarity measure of the step (b) between the respective set of samples cast from the pixels and their neighbors;
- wherein a color histogram is associated with each pixel containing the colors of all rays that hit the pixel;
- wherein the sample data sets Sx and Sy cast from pixels x and y, with respective hx k and hyk binned histograms with nb bins, are compared using the Chi-Square distance, namely
- d
- χ
- 2
-
- (
- S
- x
- ,
- S
- y
- )
- =
- ∑
- k
- =
- 1
- n
- b
-
- (
- n
- y
- n
- x
-
- h
- x
- k
- -
- n
- x
- n
- y
-
- h
- y
- k
- )
- 2
- h
- x
- k
- +
- h
- y
- k
- where ny=Σhyk and nx=Σhxk are the total number of samples in each set;
- and
- wherein, in order to take into account spatial coherence, the previous distance can be extended to patches as follows,
- d
- χ
- 2
-
- (
- P
- x
- ,
- P
- y
- )
- =
- ∑
-
- z
-
- ≤
- w
-
- d
- χ
- 2
-
- (
- S
- x
- +
- z
- ,
- S
- y
- +
- z
- )
- .
An image rendering method comprising the steps of:
- (a) performing, using a processor, a noisy estimation of a given view of a given scene of an image using a Monte-Carlo stochastic renderer, wherein general information of each ray sample is stored,
- (b) calculating a similarity measure between two sets of samples cast from two pixels of said scene and their neighbors, and
- (c) generating an output image wherein each pixel color is a weighted combination of the pixel colors generated from the step (a) with weights based on the similarity measure of the step (b) between the respective set of samples cast from the pixels and their neighbors
- the method further comprising generating an output image where each pixel color is a weighted combination of the pixel colors generated from the noisy estimation of the step (a) with weights based on the similarity measure of the step (b) between the respective set of samples cast from the pixels and its neighbors, wherein the final pixel is calculated as
- u
- _
-
- (
- x
- )
- =
- 1
- ∑
- y
-
- w
-
- (
- x
- ,
- y
- )
-
- ∑
- y
-
- u
-
- (
- y
- )
-
- w
-
- (
- x
- ,
- y
- )
- where the color value u(y) at pixel y is calculated as a weighted average of the rays in the pixel y and its neighbors by using a box, Gaussian, Mitchell or any other classical linear filter, where u denotes the color image and the filtered output color image.
Dependencies are inferred from the claim text.