Side-window Median Filter

C++ implementation of the paper Side Window Filtering(CVPR 2019) of the median filter. Successfully build on linux and android. No need to install opencv, we use the stb lib to read and save images.

Some results

Side window median filter can preserve edges and corners better than the original median filter, as shown in the following figures.

filter normal image

filter image with salt pepper noise

Build

Build linux

1. Install cmake 3.11

2. Build the project by running the scripts/build_linux.sh script and remember to modify the following parameter.

CMAKE_PATH=cmake

1. After successfully building the project, the executable runUnitTests will be executed. If every thing goes right, you should see the following outputs on your terminal.

[==========] Running 2 tests from 1 test suite.
[----------] Global test environment set-up.
[----------] 2 tests from netTest
[ RUN      ] netTest.medianFilterSpeedTest
          [0] MedianFilfer Cost time: 274.514ms
          [1] MedianFilfer Cost time: 268.929ms
          [2] MedianFilfer Cost time: 268.607ms
          [3] MedianFilfer Cost time: 268.613ms
          [4] MedianFilfer Cost time: 268.494ms
          [5] MedianFilfer Cost time: 268.851ms
          [6] MedianFilfer Cost time: 268.598ms
          [7] MedianFilfer Cost time: 268.636ms
          [8] MedianFilfer Cost time: 268.891ms
          [9] MedianFilfer Cost time: 268.796ms

          MedianFilfer Average Cost time: 269.293ms
[       OK ] netTest.medianFilterSpeedTest (2696 ms)
[ RUN      ] netTest.sideWindowMedianFilterSpeedTest
          [0] Side Window MedianFilfer Cost time: 546.448ms
          [1] Side Window MedianFilfer Cost time: 543.731ms
          [2] Side Window MedianFilfer Cost time: 543.195ms
          [3] Side Window MedianFilfer Cost time: 544.231ms
          [4] Side Window MedianFilfer Cost time: 543.111ms
          [5] Side Window MedianFilfer Cost time: 544.559ms
          [6] Side Window MedianFilfer Cost time: 544.929ms
          [7] Side Window MedianFilfer Cost time: 543.785ms
          [8] Side Window MedianFilfer Cost time: 542.317ms
          [9] Side Window MedianFilfer Cost time: 542.43ms

          Side Window MedianFilfer Average Cost time: 543.874ms
[       OK ] netTest.sideWindowMedianFilterSpeedTest (5443 ms)
[----------] 2 tests from netTest (8139 ms total)

[----------] Global test environment tear-down
[==========] 2 tests from 1 test suite ran. (8139 ms total)
[  PASSED  ] 2 tests.

Build android

1. Install adb tool with following command

sudo apt update
sudo apt install android-tools-adb android-tools-fastboot

1. Download android ndk 15c

2. Install cmake 3.11

3. Build the project by running the scripts/build_android.sh script and remember to modify the following parameter.

BUILD_ANDROID_NDK_HOME=/mnt/d/downloads/android-ndk-r15c
DEPLOY_DIR=/data/local/tmp/ldp
CMAKE_PATH=/mnt/d/downloads/cmake-3.11.4/bin/cmake

1. After successfully building the project, the executable runUnitTests will be pushed to the folder ${DEPLOY_DIR} on your android mobile and run. If every thing goes right, you should see the similar outputs on your terminal as you build on linux.

[==========] Running 2 tests from 1 test suite.
[----------] Global test environment set-up.
[----------] 2 tests from netTest
[ RUN      ] netTest.medianFilterSpeedTest
          [0] MedianFilfer Cost time: 232.383ms
          [1] MedianFilfer Cost time: 175.886ms
          [2] MedianFilfer Cost time: 173.125ms
          [3] MedianFilfer Cost time: 172.96ms
          [4] MedianFilfer Cost time: 173.178ms
          [5] MedianFilfer Cost time: 173.366ms
          [6] MedianFilfer Cost time: 173.105ms
          [7] MedianFilfer Cost time: 173.318ms
          [8] MedianFilfer Cost time: 173.126ms
          [9] MedianFilfer Cost time: 172.927ms

          MedianFilfer Average Cost time: 179.337ms
[       OK ] netTest.medianFilterSpeedTest (1805 ms)
[ RUN      ] netTest.sideWindowMedianFilterSpeedTest
          [0] Side Window MedianFilfer Cost time: 404.822ms
          [1] Side Window MedianFilfer Cost time: 404.55ms
          [2] Side Window MedianFilfer Cost time: 404.259ms
          [3] Side Window MedianFilfer Cost time: 404.199ms
          [4] Side Window MedianFilfer Cost time: 403.188ms
          [5] Side Window MedianFilfer Cost time: 404.102ms
          [6] Side Window MedianFilfer Cost time: 404.603ms
          [7] Side Window MedianFilfer Cost time: 403.546ms
          [8] Side Window MedianFilfer Cost time: 402.466ms
          [9] Side Window MedianFilfer Cost time: 403.791ms

          Side Window MedianFilfer Average Cost time: 403.953ms
[       OK ] netTest.sideWindowMedianFilterSpeedTest (4050 ms)
[----------] 2 tests from netTest (5855 ms total)

[----------] Global test environment tear-down
[==========] 2 tests from 1 test suite ran. (5855 ms total)
[  PASSED  ] 2 tests.

Run comparation on Linux

After successfully building the project on linux, you can then use the scripts/run_compare.sh to run the comparation with you own images. And remember to modify the following one or more parameters, then have fun :)!

INPUT_PATH=${basepath}/images/jian20_small_salt_pepper_noise.jpg
FILTER_RADIUS=3
ITERATION_NUM=1
MEDIAN_FILTER_OUT_PATH=${basepath}/results/median_filter_radius${FILTER_RADIUS}_iter${ITERATION_NUM}.jpg
SIDE_WINDOW_MEDIAN_FILTER_OUT_PATH=${basepath}/results/side_window_median_filter_radius${FILTER_RADIUS}_iter${ITERATION_NUM}.jpg

Run comparation on Android

After successfully building the project on android, you can then use the scripts/run_compare_android.sh to run the comparation with you own images on your android device. And remember to modify the following one or more parameters, then have fun :)!

FILTER_RADIUS=3
ITERATION_NUM=10

INPUT_NAME=jian20_salt_pepper_noise.jpg

MEDIAN_FILTER_OUT_NAME=median_filter_radius${FILTER_RADIUS}_iter${ITERATION_NUM}.jpg
SIDE_WINDOW_MEDIAN_FILTER_OUT_NAME=side_window_median_filter_radius${FILTER_RADIUS}_iter${ITERATION_NUM}.jpg

INPUT_PATH=${basepath}/images/${INPUT_NAME}
OUTPUT_PATH=${basepath}/results/