Image Convolution

Branches

• main - initial commit
• task1 - sequential grayscale bitmap convolution
• task2 - parallel convolution, kernel composition, benchmarks, CSV tables, SVG plots

Dependencies

• JDK 23
• Gradle wrapper 8.10
• Kotlin 2.1.10
• OpenCV for JVM: org.openpnp:opencv:4.9.0-0
• For plots in task2: Python 3, numpy, matplotlib

Kernels

• identity3
• box3
• gaussian3
• sharpen3

Input images are read as grayscale through OpenCV. .bmp is the intended format for the task.

Task 1

Switch to the branch:

git checkout task1

Run tests:

./gradlew test

Run sequential convolution:

./gradlew run --args="image.bmp output/result.bmp gaussian3"
./gradlew run --args="image.bmp output/sharpen.bmp sharpen3"

Task 2

Switch to the branch:

git checkout task2

Run tests:

./gradlew test

Run sequential mode:

./gradlew run --args="image.bmp output/sequential.bmp gaussian3 --mode=sequential"

Run parallel mode:

./gradlew run --args="image.bmp output/rows.bmp gaussian3 --strategy=rows --threads=8"
./gradlew run --args="image.bmp output/columns.bmp gaussian3 --strategy=columns --threads=8"
./gradlew run --args="image.bmp output/pixels.bmp gaussian3 --strategy=pixels --threads=8"
./gradlew run --args="image.bmp output/grid.bmp gaussian3 --strategy=grid --grid=4x4 --threads=8"
./gradlew run --args="image.bmp output/tile.bmp gaussian3 --strategy=grid --tile=64x64 --threads=8"

Run a filter pipeline:

./gradlew run --args="image.bmp output/pipeline.bmp gaussian3,sharpen3"

Run mathematical kernel composition:

./gradlew run --args="image.bmp output/composed-kernel.bmp gaussian3,sharpen3 --compose-kernels=true"

Benchmark Tables And Graphs

Install Python dependencies once:

python3 -m pip install numpy matplotlib

Run the full benchmark:

./scripts/run_experiment.sh image.bmp

Or run benchmark and plotting separately:

./gradlew benchmarkConvolution --args="image.bmp benchmark-results/results.csv"
python3 scripts/plot_benchmarks.py benchmark-results/results.csv plots/

Benchmark results are written as a CSV table. The main time and throughput columns are:

• avg_ms
• min_ms
• max_ms
• stddev_ms
• ci95_low_ms
• ci95_high_ms
• throughput_mp_s
• throughput_ci95_low_mp_s
• throughput_ci95_high_mp_s

The plotting script generates SVG graphs for each kernel pipeline:

• <kernel-set>-avg-ms.svg
• <kernel-set>-throughput.svg