Distort

The Distort transformation lets you distort an image by specifying a set of control points in the input image and where they are mapped to in the output image. The input image is stretched as though on a rubber sheet to move the control points to their new locations.

Control Points

Control points are displayed as an overlay on the input image. Each control point has two locations -- the first is displayed as a small circle and identifies a point in the input image. The second location specifies where you want the first location to end up in the output image. Leaving the two locations the same indicates that you do not want that part of the input image to move.

Initially, there are four control points -- one in each corner of the input image. These control points serve to lock down the corners of the image, but you can remove or adjust them if you wish.

To create a new control point, shift-click on the input image. This creates a control point both of whose locations are the same. To set the output image location of a control point, drag the control point -- this extends an arrow from the input image location to the output image location with the arrowhead at the output image location. If the output location is the same as the input location, only a small circle is displayed.

To modify a control point location, drag one end of the arrow or the other.

To remove a control point, ctrl-click on it. You cannot remove the last control point. If there is only one control point, the output image is simply a shifted version of the input image.

Method

Given a set of control points, there are many different ways to distort the image while still aligning the output locations with the input locations. The Distort transformation supports several methods which each produce different results. Except for the Shepard method, all the others distort the image using a technique called radial basis functions. Each method uses a different basis function and thus produces a different result. The Sinc method produces wilder results than the others and is provided for experimental purposes.

Shepard

Hardy

Thin Plate

Gaussian

Sinc

Inverse Multiquadratic

If you select the Hardy, Inverse Multiquadratic, Gaussian or Sinc methods, an R slider is displayed, providing you with an additional parameter that alters the distortion, generally making the effects more or less local.

Tips

While distortions of faces, bodies or architectural scenes are usually very obvious, landscapes can often be distorted significantly without looking unnatural, particularly if the horizon line is not visible.