luismavs/focusdetection

Focus area detection in photos

Focus Detection

Detects areas in focus in photos, by searching for sharpest areas and then enlarging them. Focus Detection works by converting the ouput of an high-pass filter to a mask.

To install

At the moment, the only dependencies are numpy and opencv. To install the package just do

pip install .

To run

Output is a file with same extension as the original file with a name suffix '_focus'

### run on a single image
python process.py -i input_image.jpg

### run on a directory of images
python process.py -i input_directory/ 

Parameters

The algorithm used for detection is still under development. At the moment there are two main free parameters:

• high_pass_size sets the Gaussian Blur kernel size of the High Pass Filter (strength of the high-pass filter). Higher values lead to a more aggressive separation between in-focus and out-of-focus areas.

• in_focus_regions sets how many countours the recursive median blur filter should stop, ie. how many disjoint in-focus areas we want to allow in the final mask. Values of 1 and 2 may lead to over-application of median blur recursively and shrinking of focus areas found.

Experiments on 20MP images suggest values of high_pass_size=12 and in_focus_regions=3.

Next Steps

• Build an evaluation dataset, including images with different resolution
• Train the algorithms by optimising the parameters.
• Test different kernels for the initial step.

Inspiration

BurDetection2 by Will Brennan.

Demo