The Resample function changes the raster pixel size, the resampling type, or both.
Before combining and analyzing rasters with different resolutions and map projections, it is often desirable to resample the data to a common resolution and projection.
Reprojecting an image to a different coordinate system creates an image pixel grid on an alignment other than the original image. A value for each pixel in the new image must be computed by sampling or interpolating over a neighborhood of pixels from the corresponding position in the original image.
See Cell size and resampling in analysis for more information about image resampling.
This function is useful for converting raster datasets, raster products, and mosaic datasets to a common pixel size, projection, or resampling type.
The appropriate resampling method depends on the raster data type and purpose of the resulting raster dataset as follows:
Nearest Neighbor—Used primarily for discrete data, such as a land-use classification, since it does not create new pixel values. This method is also appropriate for continuous data when you want to preserve the original reflectance values in imagery for accurate multispectral analysis. It is the most efficient in terms of processing time but may introduce small positional errors in the output image. The output image may be offset by up to half a pixel, which may cause the image to have discontinuities and a jagged appearance.
- Bilinear Interpolation and Bilinear Interpolation Plus—These methods are most appropriate for continuous data. The output image is smoother in appearance than the result of Nearest Neighbor, but the reflectance values are altered, which results in blurring or loss of image resolution. Use Bilinear Interpolation Plus for tiled data because the edges will match better. These methods are relatively computationally efficient to process.
Cubic Convolution—Suitable for continuous data. The result is geometrically less distorted than the raster achieved with Nearest Neighbor, and sharper than Bilinear Interpolation. In some cases, it can result in output pixel values outside the range of input cell values. If this is unacceptable, use the Bilinear Interpolation method instead. Cubic Convolution is computationally intensive and takes longer to process.
Majority—Suitable for discreet data, it results in a smoother appearance than the Nearest Neighbor resampling method. This method determines the value of the pixel based on the most frequentvalue in the filter window.
- Gauss Blur and Gauss Blur Plus—Appropriate for both discrete and continuous data. These methods are effective in reducing noise in resampled radar and SAR imagery that is affected by speckling. These methods are also appropriate for reducing noise and artifacts in raster data that is being down-sampled to a larger pixel size. Use Gauss Blur Plus for tiled data for better edge matching.
Average—Suitable for continuous data, it produces a smoother output image than the Nearest Neighbor resampling method.
Minimum—Appropriate for continuous data, it produces a smoother output image than the Nearest Neighbor resampling method.
Maximum—Appropriate for continuous data, it produces a smoother output image than the Nearest Neighbor resampling method.
Vector Average—Used for resampling multidimensional magnitude-direction data only.
The input pixel cell size can be different than the source pixel cell size.
The raster product to be resampled.
The pixel cell size of the input raster, which can be different from the source pixel cell size.
The pixel cell size of the output raster. The cell size can be changed, but the extent of the raster dataset will remain the same. Resampling from a user-defined cell size can be much slower than the system default, since the system default resampling processes the minimum amount of data from the closest possible display resolution.