Image_conv – Kotlin Image Convolution

Project Overview

Image_conv is a Kotlin-based image convolution project that demonstrates and benchmarks multiple convolution strategies using OpenCV.
It applies various 3×3 and 5×5 filters (e.g., blur, Gaussian blur, sharpen, edge detection, etc.) to images using both sequential and parallel approaches.

Quick Start

Build the Project

Make sure you have Java (JDK 8 or higher) and Gradle installed.

git clone https://github.com/SecretPersona5/Image_conv.git
cd Image_conv
./gradlew build

I would recommend simply shift + f10 to launch the program

This will compile the Kotlin source and download the required OpenCV native libraries.

---

Run Interactively

Launch the program in interactive mode (with a console menu):

./gradlew run
# or:
./gradlew run --args="-i"

In this mode, you will be prompted to:

• Select an input image
• Choose a convolution mode (seq, row, col, grid, pix)
• Pick a filter (e.g., blur_3x3, sharpen, etc.)
• Optionally process a whole directory (pipeline mode)

---

Process a Single Image via CLI

Run a single image with a specific mode:

./gradlew run --args="<mode> <inputImagePath> [gridBlockSize]"

Example (row-parallel convolution):

./gradlew run --args="row path/to/image.jpg"

For grid mode, add a block size:

./gradlew run --args="grid path/to/image.jpg 64"

---

Pipeline Mode (Batch Processing)

Process all images in a directory concurrently:

./gradlew run --args="pipe <inputDir> <outputDir> <mode> [blockSize] [convWorkers] [saveWorkers] [capacity]"

Example:

./gradlew run --args="pipe ./photos ./photos_out grid 128 8 1 8"

Output images will be saved in <outputDir>.

---

Benchmarks

Microbenchmark (JMH)

Run quick JMH benchmarks:

./gradlew jmhQuick

Results are saved to:

build/bench/jmh_quick.csv

It measures performance for different:

• Modes (seq, row, col, grid, pix)
• Filters (e.g., gaussian_blur_3x3)
• Image sizes
• Thread counts

---

Macrobenchmark (Pipeline)

Run full pipeline benchmarks:

./gradlew run --args='bench src/main/resources/img out/benchmarks --modes=row,seq,pix,grid,col --filters=all --tag=try1 --repeats=1 --cleanup'

Results are saved to:

out/benchmarks/

Each CSV contains throughput and timing results for all strategies across multiple runs.

---

Project Structure

src/main/kotlin/
  Main.kt              # Entry point
  Sequential.kt        # Sequential convolution
  RowParallel.kt       # Row-parallel convolution
  ColParallel.kt       # Column-parallel convolution
  GridParallel.kt      # Grid-based parallel convolution
  PixelParallel.kt     # Pixel-level parallel convolution
  Pipeline.kt          # Batch processing pipeline
  Filters.kt           # Filter definitions
  Logger.kt            # Timing logger

---

Requirements

• JDK 8+
• Gradle (wrapper included)
• OpenCV (downloaded automatically via nu.pattern.OpenCV)

---

License

MIT