Geometry Transforms

Routines manipulating an image geometry. More...

Modules
	ResizeSqrPixel
	ResizeSqrPixel functions attempt to choose source pixels that would approximately represent the center of the destination pixels.
	Resize
	Resize functions use scale factor automatically determined by the width and height ratios of input and output Region-of-Interest (ROI).
	ResizeBatch
	ResizeBatch functions use scale factor automatically determined by the width and height ratios for each pair of input / output images in provided batches.
	Remap
	Routines providing remap functionality.
	Rotate
	Rotates an image around the origin (0,0) and then shifts it.
	Mirror
	Mirrors images horizontally, vertically or diagonally.
	MirrorBatch
	Mirrors batches of images horizontally, vertically or diagonally.
	Affine Transforms
	The set of affine transform functions available in the library.
	Perspective Transform
	The set of perspective transform functions available in the library.

Detailed Description

Routines manipulating an image geometry.

These functions can be found in the nppig library. Linking to only the sub-libraries that you use can significantly save link time, application load time, and CUDA runtime startup time when using dynamic libraries.

Geometric Transform API Specifics

This section covers some of the unique API features common to the geometric transform primitives.

Geometric Transforms and ROIs

Geometric transforms operate on source and destination ROIs. The way these ROIs affect the processing of pixels differs from other (non geometric) image-processing primitives: Only pixels in the intersection of the destination ROI and the transformed source ROI are being processed.

The typical processing proceedes as follows:

1. Transform the rectangular source ROI (given in source image coordinates) into the destination image space. This yields a quadrilateral.
2. Write only pixels in the intersection of the transformed source ROI and the destination ROI.

Pixel Interpolation

The majority of image geometry transform operation need to perform a resampling of the source image as source and destination pixels are not coincident.

NPP supports the following pixel inerpolation modes (in order from fastest to slowest and lowest to highest quality):

• nearest neighbor
• linear interpolation
• cubic convolution
• supersampling
• interpolation using Lanczos window function