The image resampling process creates a new image, pixel by pixel, by establishing an appropriate value for each pixel. The resulting image is similar to the one from which it is calculated, with a very close radiometry and with a slightly different geometry.

There are two steps in the image resampling process:

• For each pixel of the output image, the first step is to find the corresponding point in the input image by passing it through the established transformation model.

• The second step is to calculate the value of the output pixel by using the value of one or several pixels located on or around the corresponding point in the input image.

The Nearest Neighbor algorithm uses only one input pixel. Bilinear Interpolation uses the four nearest neighbors and calculates the weighted mean. Cubic Convolution uses a four by four matrix of pixels and calculates a type of weighted mean.

The Nearest Neighbor algorithm is faster, but it can cause jagged edges or lines on the output image. The Cubic Convolution algorithm takes longer, but the result is much smoother. The Bilinear Interpolation is a compromise between the other two methods in both speed and image quality.

The image resampling process can be independent of the process of building a model with control points. It is used with various other kinds of models such as translations, rotations, and scale changes, or even with a neutral model used only to cut images into pieces or to change the resolution of images. A set of standard models is provided with Bentley Descartes.

Image Resampling can also be done dynamically by choosing to use the transformation matrix of the image. This process is achieved simply by turning OFF the Create a New Image option from the Register dialog.