This is an UNIX platform oriented version of Salient Region Detection and Segmentation technique represented in:

R. Achanta, S. Hemami, F. Estrada and S. Süsstrunk, Frequency-tuned Salient Region Detection, IEEE International Conference on Computer Vision and Pattern Recognition (CVPR 2009), pp. 1597 - 1604, 2009.

Check also the webpage for the paper. Most of the code is unaltered from the C++ sources given on the webpage.

The differences are:

• Use of CImg to handle picture loading and saving for platform independence
• Changed the code to work as a command line tool
• Added Python wrapper (with ctypes) for main functionality

• CImg: For loading and saving pictures.
• boost: Used for filepath manipulation. In case you don't want to use boost, you can alter the source of CImgHandler::SavePictures
• Image libraries: Defaults are libpng and libjpeg. Check CImg documentation for full list of supported formats.

The conversion has been done with minimum amount of work so any input and/or change suggestions are welcome.

1. As Command line tool:

   1.1 Compile program with make

   1.2 Call the program created:

   $>./saliency [-s] [-o=path/to/outfolder/] path/to/pic1 path/to/pic2 path/to/pic3 ...

   where optional '-s' argument tells the program to not only do saliency mapping, but also mean shift based processing and extracting the most salient objects from the pictures.

2. From Python:

   2.1 Compile program as shared library with make shared_osx or make shared_linux

   2.2 Instantiate SDRWrapper object in Python and call it:

   from sdrWrapper import SDRWrapper
   sdrw = SDRWrapper()
   doSegmentation = True # Extract most salient objects
   sdrw.saliency("path/to/outfolder", "path/to/my/pic", doSegmentation)

Created pictures are saved with different suffixes. For example, running code for bird.jpg will create three different files to the output folder when most salient objects are extracted:

• bird_sm.jpg - saliency map
• bird_ms.jpg - mean shift based segmentation
• bird_so.jpg - most salient object