VXLIB Function Reference

---

Contents

1. Introduction
2. Notational Conventions
3. OpenVX 1.1 Vision Functions
   1. Absolute Difference
   2. Accumulate
   3. Accumulate Squared
   4. Accumulate Weighted
   5. Arithmetic Addition
   6. Arithmetic Subtraction
   7. Bitwise AND
   8. Bitwise EXCLUSIVE OR
   9. Bitwise INCLUSIVE OR
   10. Bitwise NOT
   11. Box Filter
   12. Canny Edge Detector
   13. Channel Combine
   14. Channel Extract
   15. Color Convert
   16. Convert Bit depth
   17. Custom Convolution
   18. Dilate Image
   19. Equalize Histogram
   20. Erode Image
   21. Fast Corners
   22. Gaussian Filter
   23. Non Linear Filter
   24. Harris Corners
   25. Histogram
   26. Gaussian Image Pyramid
   27. Laplacian Image Pyramid
   28. Reconstruction from a Laplacian Image Pyramid
   29. Integral Image
   30. Magnitude
   31. Mean and Standard Deviation
   32. Median Filter
   33. Min, Max Location
   34. Optical Flow Pyramid (LK)
   35. Phase
   36. Pixel-wise Multiplication
   37. Remap
   38. Scale Image
   39. Sobel 3x3
   40. TableLookup
   41. Thresholding
   42. Warp Affine
   43. Warp Perspective

---

Introduction

VXLIB provides a collection of C-callable high-performance routines that are developed to be compatible with the kernels defined in the Khronos OpenVX specificaiton. These functions exploit the high performance capabilities of Texas Instruments DSPs.

All functions in VXLIB have been developed for a little-endian memory model. Some may work in a big-endian memory model as well, however, their functionality is not guaranteed.

---

Notational Conventions

All functions within VXLIB following the naming conventions described below.

1. Each function begins with the prefix VXLIB_. This serves to prevent namespace conflicts.
2. The prefix is followed by an English functional descriptor. The functional descriptor was chosen to follow the descriptor used in OpenVX node API, where applicable.
3. Further description after the English words (i.e. variants, border modes, bit depths, numbers) are separated by underscores: '_'
4. All kernels use UNDEFINED border modes, unless specified by the following convention:
   bc=constant border mode
   br=replicate border mode
5. Filter kernel or window sizes are MxN format (e.g. 3x3).
6. Conversion of channels/bit depths are XtoY format (e.g. 4to1, 8to16).
7. Data types of buffer parameters come last in the kernel name, and use the following codes:
   i=input
   o=output
   io=inout
   c=coefficients
   s=signed
   u=unsigned
   f=float
8. Scalar inputs/outputs, lists, and scratch buffers are not part of the name.