Efficient image transformation

Patent metadata

Field	Value
Publication number	US-2016292832-A1
Application number	US-201514678299-A
Country	US
Kind code	A1
Filing date	Apr 3, 2015
Priority date	Apr 3, 2015
Publication date	Oct 6, 2016
Grant date	—

Abstract

The present disclosure provides a multi-stage image mapping mechanism for mapping a distorted image to a rectified image. For example, the multi-stage image mapping mechanism can remove homography from a distorted image to reconstruct a rectified image in two-stages: (1) a first stage in which distortion is partially removed from a distorted image to generate an intermediate image, and (2) a second stage in which residual distortion is removed from the intermediate image to recover the rectified image.

First claim

1 . A machine vision system comprising:
a processor configured to run a computer program stored in memory that is configured to cause the processor to:
receive a first image and a first transform associated with the first image, wherein the first transform relates the first image to a second image, and is constrained to map a plurality of pixels along a dimension of the second image to a plurality of locations along a dimension of the first image;
determine, for a first pixel of the plurality of pixels, a coordinate of a corresponding first location in the first image;
select one of a plurality of kernels based on a sub-pixel phase associated with the coordinate of the first location, wherein the sub-pixel phase is indicative of a sub-pixel offset of the first location from a pixel grid of the first image; and
generate the second image from the first image by, in part, applying the selected kernel on at least one pixel around the first location in the first image to determine an image value for the first pixel in the second image.

2 . The machine vision system of claim 1 , wherein the processor is configured to determine the coordinate of the first location by operating the first transform on a coordinate of the first pixel.

3 . The machine vision system of claim 1 , wherein the processor is configured to determine, for a second pixel of the plurality of pixels, a coordinate of a corresponding second location in the first image by adding a constant to the coordinate of the first location.

4 . The machine vision system of claim 3 , wherein the processor is configured to:
select a second one of the plurality of kernels based on a sub-pixel phase associated with the coordinate of the second location; and generate the second image from the first image by, in part, applying the second one of the plurality of kernels on at least one pixel around the second location in the first image to determine an image value for the second pixel in the second image

5 . The machine vision system of claim 1 , wherein the plurality of kernels correspond to an identical function that is sampled at a plurality of sub-phases.

6 . The machine vision system of claim 1 , wherein the processor is configured to:
receive a second transform associated with the second image, wherein the second transform relates the second image to a third image, and is constrained to map a plurality of pixels along a dimension of the third image to a plurality of locations along the dimension of the second image; determine, for a first pixel of the third image, a coordinate of a corresponding first location of the third image; select one of the plurality of kernels based on a sub-pixel phase associated with the coordinate of the first location in the third image; and generate the third image from the second image by, in part, applying the one of the plurality of kernels on at least one pixel around the first location of the third image to determine an image value for the first pixel of the third image.

7 . The machine vision system of claim 6 , wherein the first transform and the second transform collectively represent an affine transformation.

8 . The machine vision system of claim 1 , wherein the first transform comprises a down-sampling operation.

9 . The machine vision system of claim 1 , wherein the first transform comprises a first transformation matrix.

10 . A computerized method comprising:
receiving a first image and a first transform associated with the first image, wherein the first transform relates the first image to a second image, and is constrained to map a plurality of pixels along a dimension of the second image to a plurality of locations along a dimension of the first image; determining, for a first pixel of the plurality of pixels, a coordinate of a corresponding first location in the first image; selecting one of a plurality of kernels based on a sub-pixel phase associated with the coordinate of the first location, wherein the sub-pixel phase is indicative of a sub-pixel offset of the first location from a pixel grid of the first image; and generating the second image from the first image by, in part, applying the selected kernel on at least one pixel around the first location in the first image to determine an image value for the first pixel in the second image.

11 . The computerized method of claim 10 , wherein determining the coordinate of the first location comprises operating the first transform on a coordinate of the first pixel.

12 . The computerized method of claim 10 , further comprising determining, for a second pixel of the plurality of pixels, a coordinate of a corresponding second location in the first image by adding a constant to the coordinate of the first location.

13 . The computerized method of claim 12 , further comprising:
selecting a second one of the plurality of kernels based on a sub-pixel phase associated with the coordinate of the second location; and generating the second image from the first image by, in part, applying the second one of the plurality of kernels on at least one pixel around the second location in the first image to determine an image value for the second pixel in the second image

14 . The computerized method of claim 10 , further comprising:
receiving a second transform associated with the second image, wherein the second transform relates the second image to a third image, and is constrained to map a plurality of pixels along a dimension of the third image to a plurality of locations along the dimension of the second image; determining, for a first pixel of the third image, a coordinate of a corresponding first location of the third image; selecting one of the plurality of kernels based on a sub-pixel phase associated with the coordinate of the first location in the third image; and generating the third image from the second image by, in part, applying the one of the plurality of kernels on at least one pixel around the first location of the third image to determine an image value for the first pixel of the third image.

15 . The computerized method of claim 14 , wherein the first transform and the second transform collectively represent an affine transformation.

16 . The computerized method of claim 10 , wherein the first transform comprises a down-sampling operation.

17 . The computerized method of claim 10 , wherein the first transform comprises a first transformation matrix.

18 . A non-transitory computer readable medium having executable instructions associated with a homography decomposition module and a homography removal module, operable to cause a machine vision system to:
receive a first image and a first transform associated with the first image, wherein the first transform relates the first image to a second image, and is constrained to map a plurality of pixels along a dimension of the second image to a plurality of locations along a dimension of the first image; determine, for a first pixel of the plurality of pixels, a coordinate of a corresponding first location in the first image; select one of a plurality of kernels based on a sub-pixel phase associated with the coordinate of the first location, wherein the sub-pixel phase is indicative of a sub-pixel offset of the first location from a pixel grid of the first image; and generate the second image from the first image by, in part, applying the selected kernel on at least one pixel around the first location in the first image to determine an image value for the first pixel in the second image.

19 . The non-

Assignees

• Cognex Corp

Inventors

• Bogan Nathaniel

Classifications

• G06T2207/20221

  Image fusion; Image merging · CPC title

• G06T5/50

  using two or more images, e.g. averaging or subtraction · CPC title

• G06T3/00 Primary

  Geometric image transformations in the plane of the image · CPC title

• G06T3/0075

  Physics · mapped topic

• G06T5/006 Primary

  Physics · mapped topic