The Intel^® Image Processing Library provides a set of low-level image manipulation functions in standard DLLs and static libraries form. The functions are optimized for Intel^® Architecture processors, and are particularly effective at taking advantage of MMX™ technology, the Streaming SIMD Extensions (SSE) and SSE-2.. Currently, versions have been developed for the Intel486™ and compatible processors, the Pentium® processor, the Pentium Pro processor, the Pentium processor with MMX technology, the Pentium II processor, the Pentium III processor, and, most recently, the Pentium 4 processor.  A separate DLL is available for each processor.

The library contains functions that perform filtering, thresholding, and transforms (FFT, DCT, geometric), as well as arithmetic and morphological operations. The library uses a flexible image format, supporting channels of 1-, 8-, 16-, and 32-bit integer pixels, as well as 32-bit floating point pixels, with an arbitrary number of channels per image. Conversion to and from the Windows* DIB (device independent bitmap) image format is supported, as are conversions between color and gray-scale.

Major changes from the Image Processing Library 1.0 include many more geometric functions, mask region of interest, floating point support, example code, and color space conversions.

Changes since version 2.2  include further optimization for Pentium III processors and support for the Pentium 4 processor.

Additional information on this software as well as other libraries is available at the Intel Performance Libraries homepage.

The Intel^® Image Processing Library requires the following software and hardware:

Software Requirements: Windows* 95, 98, 2000, or Windows NT* 3.51 or 4.0.

Intel^® C/C++ Compiler, Borland* C/C++ 5, or Microsoft* Visual C/C++* 4.2 or later. Other compatible compilers have not been tested extensively but might work.

You can install the Image Processing Library package by downloading and running the appropriate self-extracting installation files from the Web site.  See the  Intel Image Processing Library homepage for updates, when available.

To get started using the library, the following steps are required:

• Include the file "ipl.h" in your program.
• Call Image Processing Library functions.
• Link to the file ipl.lib in the lib\msvc or lib\bcc directory.
• Put the DLLs ipl.dll, cpuinf32.dll, and iplxxx.dll in your path before executing your program.

• The "waterfall model" of DLL selection was implemented; now the dispatcher ipl.dll will search for the best available DLL for the detected processor.
• Channel of Interest now takes precedence over alpha channel for arithmetic, logical, morphological, histogram, geometric, and linear transform functions, and convolution, set, copy, and threshold.
• A bug that caused iplDecimate and iplResize to request out-of-bounds tiles was fixed.
• Bugs were fixed in iplMpyRCPack2D, so that this function is now available for use.
• Some new functions: iplSetPixel, iplGetPixel, iplConvertFromDIBSep, iplCloneImage, and iplResize. These functions are primarily derived from other library functions but provide a more convenient interface.
• Delphi examples. The library now includes a .pas file declaring library functions for use with Delphi, and binary executable and source code for a simple image editor. Please note that while we hope these files will be helpful, this interface is not as extensively tested as the remainder of the library.
• A tutorial is available. This tutorial is HTML based and requires Internet Explorer 3.0 or later to run the examples from within the browser or Netscape 3.0 to view the explanations. This tutorial includes several example routines that should be generally useful, such as DIB loading, image display, and timer functions.
• Source code for video effects is provided. Building video effects, including Microsoft DirectShow* and Adobe* Premiere* filters, is demonstrated with six examples. Pre-built versions of the binaries are also available.
• The performance of these and other functions has been significantly enhanced: iplXYZ2RGB, iplRGB2XYZ, iplYCrCb2RGB, iplRGB2YCrCb, iplConvertToDIB with palettes, iplReduceBits with dithering, iplConvolve2D in general and more with 3x3 and 5x5 kernels, iplWarpAffine with linear interpolation.
• Alpha channel interpretation by non-alpha-compositing functions has changed. In version 1.0, the behavior was not consistent. Now the uniform rule, excepting only geometric functions, is that the alpha channel is not to be operated upon by these functions. That is, the destination alpha channel will remain unchanged in all but alpha-compositing and geometric functions.
• The following bugs have been resolved:
• YCrCb – color plane order changed from YCbCr to YCrCb
• Small issues fixed in: Open, Close, PerspectiveWarp, MaxFilter, Convolve2D, FixedFilter, Mirror, Norm
• ipliedit.exe supports Abs, AlphaCompositeC, and PremultiplyAlpha