Images contrast adjuster

Parallel contrast adjuster using OpenMP library.

Example

What's happening?

The task to familiarize with parallel programming and the OpenMP library possibilities

Translate the values from the YCBCr color space to the RGB
Let change the pixel values to obtain maximum contrast without changing the color (hue). That can be achieved by linear adjusting the contrast in the luminance channel Y.

The input file contains data in PPM (P6) format.
The file doesn't contain comments. Maximum brightness value: 255

The output file will be a new image in the format PPM (P6).

Benchmark

Test using test.cmd

As we can see, the min time is reached when the number of threads is nearly equal to the number of cores. Hyper-threading is not beneficial because each thread uses the same pipelining instructions in a superscalar.

Test system: Intel Core i7-9750H, 6 cores, 12 threads

Image’s size: 5184 * 3456 pixels, 52 489 KB
Path to input image: image.ppm
Path to output image: result.ppm
Enter max threads value: 12
Path to .exe: contrast_adjuster.exe
Threads: 0, time: 182.999849 ms
Threads: 1, time: 505.999804 ms
Threads: 2, time: 239.000082 ms
Threads: 3, time: 164.999962 ms
Threads: 4, time: 149.000168 ms
Threads: 5, time: 130.000114 ms
Threads: 6, time: 135.000229 ms
Threads: 7, time: 137.000084 ms
Threads: 8, time: 137.000084 ms
Threads: 9, time: 145.999908 ms
Threads: 10, time: 139.000177 ms
Threads: 11, time: 149.999857 ms
Threads: 12, time: 141.999960 ms

Run

Clone this repo:

git clone https://github.com/Dalvikk/contrast_adjuster
cd contrast_adjuster

Compile using g++/clang++:

g++/clang++ main.cpp -std=c++14 -fopenmp -o contrast_adjuster

or using cmake:

mkdir build
cd build
cmake ..
make (or mingw32-make if you use windows and mingw)

Run:

./contrast_adjuster (or contrast_adjuster.exe) <num of threads> <input_file> <output_file>

The number of threads can be 0, in that case, OpenMP self decide how many will use

Algorithm

1. Read the image in RGB color space
2. Translate from RGB into YCbCr using these formulas from wikipedia:
3. 
   
4. Find min and max Y. Be careful to avoid race condition
5. Linear adjusting Y values from [min_Y; max_Y] to [0, 255]
6. Translate from YCbCr to RGB

Steps 2-5 can easily parallelize