Mean Filter

Introduction

Given an image, we’ll apply a mask/ kernel of order-X ( of (2x+1) x (2x+1) dimension ) on each pixel of image & keep putting mean of pixel intensities from its neighborhood, at each I(x, y).

Performance

First I was processing each image pixel concurrently. Then applying order-5 Mean Filter on an image of dimension 3200x1907, took ~2 min.

performance

Now it has been improved, processing each row of image matrix concurrently. In previous case it was not doing that much work in each worker thread, but more works thrown at thread pool. Which has been drastically reduced with introduction of row based concurrent processing. Now it took ~14s, which is almost 8x improvement.

performance

Improvement will be clearer with processing of larger images.

Testing

Performance testing done on an image of dimension 3200x1907, by applying Mean Filter of order-5.

Source	Sink

Usage

Put below code snippet in Java source file.

import in.itzmeanjan.filterit.ImportExportImage;
import in.itzmeanjan.filterit.filter.MeanFilter;


public class Main{

	public static void main(String [] args){
		System.out.println(
            ImportExportImage.exportImage(
                new MeanFilter().filter("bridge.jpg", 5), 
                "meanFiltered.jpg"));
	}

}

Compile and run project, make sure you before that you add in.itzmeanjan.filterit.jar as your project dependency.

Results

Order of Filter	Image
0
1
2
3
4
5

filterIt