Image rectification


From wiki: ‘Image rectification is a transform process used to project images onto a common plane.’

Can anyone explain it in simple terms as I fail to actually understand what they mean

In: Technology

Imagine you have two cameras. The cameras are not in the same place. But they are both looking at the same thing. When they take pictures, they will take pictures of the same thing. But the pictures will look different because the cameras are looking at it from different places.

But perhaps you want to compare the pictures of the thing. So it might be useful to manipulate the images you took in a way so that you get two manipulated images that try to show the thing as if the pictures were taken from a camera at the same position.

The distortion you need to do to these pictures to achieve that is image rectification. It involves skewing, rotation, and stretching the starting images.

There are a few types. To put it simply, if you take a picture of an object, then take another having rotated the camera slightly, then you can rectify the two images using image rectification processes. The basic process is finding groups of pixels that look similar in each image. When you find several groups that you feel confident are matches in each image, you can calculate how much each of are offset from one image to the other. Then you can take all of those offsets to calculate how much you rotated the camera from picture 1 to picture 2. Once you know the rotation you can unrotate the second image to match the first. This can be done with rotation, zoom, scale, and translation. Usually you just create a matrix to represent all of the transforms, then warp one image to the other using that transform matrix.