Skip to content
A project to make RawTherapee's processing algorithms more readily available.
C++ C CMake
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.
cmake/Modules cmake: Add cpack config Mar 20, 2019
CMakeLists.txt Default to RelWithDebInfo Jun 23, 2019
ProcessorTargets.cmake native builds, auto generate rtprocess.pc from rtprocessin.pc Sep 19, 2018


This is a project that aims to make some of RawTherapee's highly optimized raw processing routines readily available for other FOSS photo editing software.

The goal is to move certain source files from RawTherapee into this library. Thus, any changes to the source can be done here and will be used by the projects which use librtprocess.

librtprocess currently is maintained by developers of the following projects:

. Filmulator

. HDRMerge

. LuminanceHdr

. PhotoFlow

. rawproc

. RawTherapee

... the latter is where currently all the code comes from ;-)

This is version 0.11.0, which furnishes the following routines:

  • ahd_demosaic
  • amaze_demosaic
  • bayerfast_demosaic
  • dcb_demosaic
  • hphd_demosaic
  • igv_demosaic
  • lmmse_demosaic
  • rcd_demosaic
  • vng4_demosaic
  • markesteijn_demosaic
  • xtransfast_demosaic
  • CA_correct
  • HLRecovery_inpaint

Build instructions:

  1. Make a subdirectory named build, and cd to that directory.
  2. Run cmake -DCMAKE_BUILD_TYPE="Release" ..
  3. Run make
  4. Run make install as root.

Build instructions for Windows msys2 environment:

  1. Make a subdirectory named build, and cd to that directory.
  3. Run make
  4. Run make install.

Build instructions for macOS:

Prerequisites: XCode/XCode command line tools. An optional SDK (this example uses macOS 10.9). An implementation of OpenMP, for example libiomp.5.

  1. Make a subdirectory named build, and cd to that directory.
  2. On macOS 10.12 Sierra, run sudo cmake -DCMAKE_BUILD_TYPE="release" -DPROC_TARGET_NUMBER="1" -DCMAKE_C_COMPILER="clang-mp-3.9" -DCMAKE_CXX_COMPILER="clang++-mp-3.9" -DCMAKE_CXX_FLAGS=-I/opt/local/include -DCMAKE_OSX_SYSROOT="/Applications/" -DCMAKE_OSX_DEPLOYMENT_TARGET="10.9" -DOpenMP_C_FLAGS=-fopenmp="libiomp5" -DOpenMP_CXX_FLAGS=-fopenmp="libiomp5" -DOpenMP_C_LIB_NAMES="libiomp5" -DOpenMP_CXX_LIB_NAMES="libiomp5" -DOpenMP_libiomp5_LIBRARY="/opt/local" -DCMAKE_INSTALL_PREFIX=/opt/local ..

    On macOS 10.14 Mojave, run cmake -DCMAKE_BUILD_TYPE="release" -DPROC_TARGET_NUMBER="1" -DCMAKE_CXX_COMPILER="clang++" -DCMAKE_CXX_FLAGS=-I/opt/local/include -DCMAKE_OSX_SYSROOT="/Applications/" -DCMAKE_OSX_DEPLOYMENT_TARGET="10.9" -DOpenMP_CXX_FLAGS=-fopenmp=lomp -DOpenMP_CXX_LIB_NAMES="libomp" -DOpenMP_CXX_FLAGS="-Xpreprocessor -fopenmp /opt/local/lib/libomp.dylib -I/opt/local/include" -DOpenMP_CXX_LIB_NAMES="libomp" -DOpenMP_libomp_LIBRARY=/opt/local/lib/libomp.dylib -DCMAKE_INSTALL_PREFIX=/opt/local -DCMAKE_SHARED_LINKER_FLAGS=-L/opt/local/lib ..
  3. Run sudo make -j$(sysctl -n hw.ncpu) install

Optional switches to be included in the cmake command:

  1. To build in verbose mode, include -DVERBOSE=ON
  2. If you make your own builds, include -DPROC_TARGET_NUMBER=2 for maximum speed. Keep in mind that this build will only work on the machine you built it.

Using librtprocess:

Include -lrtprocess, and #include <rtprocess/librtprocess.h> to use this library.


The demosaic routines expect raw data in the form 1) single-channel, 2) float, 3) range 0.0 - 65535.0. This roughly corresponds to what the raw libraries deliver, e.g. Libraw's mosaic is single-channel unsigned short 0-65535, except for the float number format, so at least a unsigned short -> float cast is probably required.

The raw data array expected by the demosaic routines is float**, which is an array of pointers to pointers. This hierarchical pointer arrangement is "row-major", that is, the first array of pointers point to pointers that point to a contiguous block of memory containing the pixel data for a row. This choice of storage is for fast performance, as individual pixels can be accessed with pointer dereferencing, but at the expense of rather convoluted memory management in C.

The demosaic routines' output are three separate float** arrays, one for each channel. This output organization helps certain applications that like to start with the separate channels; for 'regular' use, the red, green, and blue floats have to be loaded separately into each channel of the destination RGB struct or array.

Here's a code segment that demonstrates the marshalling/demarshalling of data, taken from the rawproc application. Note: rawproc's internal image is a row-major contiguous array of RGB floats in the range 0.0 - 1.0, so the code includes the logic to convert from/to this structure.

    //build the input and output data structures. 'w' and 'h' are the image width and height.
		float **rawdata = (float **)malloc(h * sizeof(float *));
		rawdata[0] = (float *)malloc(w*h * sizeof(float));
		for (unsigned i=1; i<h; i++) 
			rawdata[i] = rawdata[i - 1] + w; 

		float **red     = (float **)malloc(h * sizeof(float *)); 
		red[0] = (float *)malloc(w*h * sizeof(float));
		for (unsigned i=1; i<h; i++) 
			red[i]     = red[i - 1] + w;

		float **green     = (float **)malloc(h * sizeof(float *)); 
		green[0] = (float *)malloc(w*h * sizeof(float));
		for (unsigned i=1; i<h; i++) 
			green[i]     = green[i - 1] + w;

		float **blue     = (float **)malloc(h * sizeof(float *)); 
		blue[0] = (float *)malloc(w*h * sizeof(float));
		for (unsigned i=1; i<h; i++) 
			blue[i]     = blue[i - 1] + w;

    //loads the internal data to the librtprocess rawData structure.  The interal data's red channel 
    //is arbitrarily chosen as a monochrome image is represented R=G=B:
		#pragma omp parallel for num_threads(threadcount)
		for (unsigned y=0; y<h; y++) {
			for (unsigned x=0; x<w; x++) {
				unsigned pos = x + y*w;
				rawdata[y][x] = image[pos].r * 65535.f;
		vng4_demosaic (w, h, rawdata, red, green, blue, cfarray, f);

    //assemble the demosaiced RGB array from the individual channels:
		#pragma omp parallel for num_threads(threadcount)	
		for (unsigned y=0; y<h; y++) {
			for (unsigned x=0; x<w; x++) {
				unsigned pos = x + y*w;
				image[pos].r = red[y][x] /65535.f;
				image[pos].g = green[y][x] /65535.f;
				image[pos].b = blue[y][x] /65535.f;
		free (blue[0]);
		free( blue );
		free (green[0]);
		free( green );
		free (red[0]);
		free( red );
		free (rawdata[0]);
		free( rawdata );

The Bayer demosaic routines also take in a cfarray parameter that is a 2x2 array corresponding to the indexing of the color filters corresponding to the top left corner of the raw image plane. The xtrans parameter for the X-Trans routines is similar but has dimensions of 6 by 6. For these, 0 corresponds to red, 1 corresponds to green channel one, 2 corresponds to blue, and 3 corresponds to green channel two. Some algorithms require both greens, others only one.

Highlight Recovery

The highlight recovery algorithm uses inpainting to reconstruct clipped highlights when not all channels are clipped. The input data should be full RGB for each pixel, in the raw color space, with the white balance multipliers already applied to it. chmax is simply the maximum pixel value in each of the three color channels. clmax is the raw clip point for each channel; that is, the whitepoint minus the blackpoint for each channel, multiplied by the white balance multipliers.

You can’t perform that action at this time.