IMG FILTER

This program performs two different image processing tasks:

• Noise removal by applying median filtering
• Sharpening images via the unsharp masking method

It makes use of Nvidia Warp API, which allows the use of CUDA-capable Nvidia GPUs to speed up processing of images.

Check if your Nvidia GPU is CUDA-capable here.

See the GPU acceleration section to see how to improve performance if you have access to Nvidia GPUs.

Dependencies

Ensure you have the following python libraries installed:

• numpy
• pillow (Pillow)
• warp

You can install them with:

pip install numpy pillow warp

Components

main.py:

• Main program that will call either the image sharpening or noise removal algorithm

sharpen.py:

• Contains the unsharp masking algorithm for sharpening images

noiseRemove.py:

• Contains the noise removal algorithm that makes use of median filtering

How to run the program

The program is executed in the command line in the following format:

python3 main.py <algType> <kernSize> <param> <inFileName> <outFileName>

Arguments

<algType>: The type of algorithm to apply:

• -n - Apply median filtering for noise removal.
• -s - Apply unsharp masking for sharpening.

<kernSize>:

• The size of the kernel (must be a positive odd integer, e.g., 3, 5, 7).

<param>: A parameter specific to the algorithm:

• For median filtering (-n):
  • this is a dummy value (just enter '1').
• For unsharp masking (-s),
  • this is the scaling factor for sharpening.

<inFileName>:

• Path to the input image file.

<outFileName>:

• Path to save the output image file (must include an image extension, e.g., .png, .jpg).

Examples

Sharpening

python3 main.py -s 3 1 image.png result.png

Noise removal

python3 main.py -n 3 1 image.png result.png

GPU acceleration

If you have Nvidia GPUs that are CUDA-capable, you may be able to speed up the program by changing a line in the code.

At line 10 in main.py, the device is originally set to:

device = "cpu"

Change "cpu" to "cuda:0".

If you have multiple CUDA-capable GPUs on your system, you may need to change the "0" to a different number.

The format should be "cuda:i".

See the Warp documentation for more details.

Troubleshooting

One issue some may run into is if the image is using unsupported channels. Make sure the image is only using the following channels:

• RGB
• RGBA
• Greyscale