Image Processing Lab

Lab exercises for the course on image processing (in French).

Organization

There are 5 files to complete, each one corresponds to a course chapter:

1. src/tpHistogram.cpp: histogram manipulation functions related course chapter
2. src/tpConnectedComponents.cpp pixel adjacency and connected components related course chapter
3. src/tpGeometry.cpp geometric image transforms related course chapter
4. src/tpConvolution.cpp linear image filters related course chapter
5. src/tpMorphology.cpp non-linear image filters related course chapter

How to

Environment

This lab requires a working *nix environment with a c++ 11 compiler and valgrind.

The library OpenCV4 is required to compile this project. To install a local version of this library just execute the script source ./init.sh. This script will automatically download the library and set up the environment variable required by the makefile.

Important: If you install OpenCV with the script source ./init.sh, you have to run the script in every terminal used to work on this project!

Compilation

Just run make in the top directory, this will create several executables in the bin directory.

Tip When using make, use the argument -j n to launch a parallel compilation on n cores. For example: make -j 4 while compile up to 4 files in parallel.

Executables

Each exercise is associated to its own command line tool. For exemple, the first function to code, inverse in the file src/tpHistogram.cpp, is associated to the command line tool bin/inverse generated by make.

The usage of the command line tools is obtained with the argument --help, for example, executing the command ./inverse --help inside the bin directory produces the following output:

Inverse
Usage: ./inverse [OPTIONS]

Options:
-h,--help                   Print this help message and exit
-I,--inputImage TEXT        Input image filename
-O,--outputImage TEXT       Output image filename
-S,--show                   Display input and output images in new windows

Unit tests

Each command line tool is associated to a unit test that can be executed with the tool bin/test. For example the tool inverse can be tested with the command ./test -P inverse. If the test detects an error, you can ask to see an error map with the argument -S: this will display an image showing where the errors are located: the brighter a pixel is, the further it is from the expected result.

Important: The unit tests work by comparing the result of your functions to results produced with reference implementations. Implementation details, especially when working with floatting point values can lead to small disprecancies between similar results. While the test program tries to deal with this issue it can still detect false positives, i.e., say that a function is wrong while it is indeed right. In case of doubt call a professor to check your work.

As with any unit test, it is also possible that the unit tests suffer from false negatives, i.e., say that a program is correct while it is not.