Random imaging
AssigneeGoodrich Corporation
InventorsHongcheng Wang, Alan Matthew Finn, Serge L. Shishkin, Sergei F. Burlatsky
Life of this patent
Estimating the term…
Abstract
An image processing system comprises a filter associated with a sensor array, which is operable to capture an image. The filter is provided to separate a plurality of distinct qualities of light from the scene to be captured. The filter has filter portions associated with the plurality of distinct qualities of light which are spatially pseudo-randomly ordered relative to each other. The image processing system also comprises an image reconstruction algorithm specifically designed to operate with the filter.
Claims15
An image processing system comprising:
- a filter associated with a sensor array, said sensor array being operable to capture an image;
- and
- said filter constructed and arranged to separate a plurality of distinct qualities from a scene to be captured, with said filter having filter portions associated with said plurality of distinct qualities which are spatially pseudo-randomly ordered relative to each other;
- an image reconstruction algorithm;
- wherein said image reconstruction algorithm jointly reconstructs one or more distinct images from a plurality of pseudo-random images associated with each of said plurality of distinct qualities;
- wherein the pseudo-random images are modeled with a physical imaging constraint by acquiring said pseudo-random images with different effective exposure times for different pixels and utilizing a camera response function to model an irradiance value is represented as:
- xi=f(log(δti))
- where δti is the effective exposure time, and f is the camera response function;
- and
- said reconstruction algorithm utilizes dictionary learning.
An image processing system comprising:
- a filter associated with a sensor array, said sensor array being operable to capture an image;
- said filter constructed and arranged to separate a plurality of distinct qualities from a scene to be captured, with said filter having filter portions associated with said plurality of distinct qualities which are spatially pseudo-randomly ordered relative to each other;
- an image reconstruction algorithm;
- wherein said reconstruction algorithm utilizes one or more of dictionary learning and matrix completion;
- and
- wherein said dictionary learning image reconstruction utilizes image patches, and pseudo-random images are divided into a set of overlapping patches, each of a size a×a wherein a is greater than 1, and a dictionary is learned from the pseudo-random images as follows:
- min
- α
- ,
- Φ
-
- ∑
- i
- =
- 1
- n
-
-
- 1
- 2
-
-
- x
- i
- -
- Φα
- i
-
- 2
- 2
- such that ∥αi∥1≦τ0
- where xi are image patches, Φ is the dictionary, αi are a sparse coefficients, and τ0 is a small constant.
A method of image reconstruction using a filter comprising the steps of:
- providing a filter used to separate a plurality of distinct qualities from a scene to be captured, with said filter having filter portions associated with said plurality of distinct qualities of a scene which are spatially pseudo-randomly ordered relative to each other, and said filter being associated with a sensor array,
- reconstructing one or more distinct images from one or more of said distinct qualities of said scene;
- and wherein said reconstruction of the one or more distinct images is done jointly from a plurality of pseudo-random images associated with each of said plurality of distinct qualities;
- and
- wherein the pseudo-random images are modeled with a physical constraint by acquiring said pseudo-random images with different effective exposure times for different pixels and utilizing a camera response function to model an irradiance value is represented as:
- xi=f(log(δti))
- where δti is the effective exposure time, and f is the camera response function.
Dependencies are inferred from the claim text.