Keyetech Performance Primitives
The Keyetech Performance Primitives (KPP) offer the probably world's fastest implementations of various image processing routines.
You can either develop your computer vision application with your library of choice and call KPP routines explicitly, or you can use the computer vision library Integrating Vision Toolkit (IVT) for implementation, which automatically integrates the KPP. By purchasing and installing the KPP on your system, your IVT applications automatically will run faster without any modifications.
The KPP offer high-end optimized implementations of various image processing routines for different platforms. Currently the KPP are available for Intel CPUs offering MMX and SSE2 (Intel Pentium 4, Intel Core 2 Duo, Intel Core 2 Quad, Intel Atom, or compatible). Some routines can optionally be executed on the GPU of NVIDIA graphic cards to achieve even further speedup. In any case, integration is simple!
In the following table, runtimes of exemplary funtions are listed. Compare yourself with the information provided by other suppliers! A demo application for measurement on your own computer system is available upon request per e-mail.
Thresholding (binarization) | 0,022 ms |
Gaussian filter 3×3 | 0,10 ms |
Gaussian filter 5×5 | 0,29 ms |
Sobel X / Sobel Y | 0,12 ms |
Sobel fused | 0,20 ms |
Binary 3×3 Erosion + Dilation | 0,15 ms |
Color segmentation in HSV color space | 0,078 ms |
RGB to HSV conversion | 1,5 ms |
Canny edge detector | 2,1 ms |
Harris corner detector | 4,2 ms |
Matching of point features | 2,3 ms* |
Runtimes using the KPP, measured for processing an image of size 640×480 on an Intel Core 2 Duo 3.0 GHz (the runtimes can vary with different input data).
* 500 against 500 features. Utilizes GPU processing.