Haskell Pics. An image processing tool written in Haskell.

A simple command line utility performing some basic operations on images.
Libraries used: JuicyPixels combined with Repa.
I made this program as a practice in Haskell and image processing. There might be some performance flaws and bugs.

REQUIREMENTS:

GHC compiler
Packages: JuicyPixels, Repa, Criterion

COMPILING:

Compile using 'make' or manually by one of the following commands:
ghc hspics.hs
ghc -O -threaded -rtsopts --make hspics.hs
The latter utilizes optimized and multithreaded array operations by Repa

USAGE:

Run the tool in command line like this:
hspics.exe [command] [parameters]
e.g: hspics.exe grayscale some_picture.jpg

For multithreading run with the following flags (recommended!):
hsimg.exe [command] [parameters] +RTS -N2 -H
where '-N2' means running on 2 cores

Different command line arguments correspond to different types of image processing.
Possible input formats: same as in JuicyPixels lib.
Output format: PNG.
The output of processing is always placed in "output.png" file.
ATTENTION: if you run program when there is already an "output.png" file, it will be overridden!

POSSIBLE COMMANDS:

Basic color transformations:

grayscale [imagePath]
only_red [imagePath]
only_blue [imagePath]
only_green [imagePath]
negative [imagePath]
sepia [imagePah]

YCbCr color coordinates:

only_y [imagePath]
only_cb [imagePath]
only_cr [imagePath]

HSL color coordinates:

only_h [imagePath] <-- uses only h, sets s=0.5 and l=0.5
only_s [imagePath] <-- uses h and s, sets l=0.5
only_l [imagePath] <-- uses only l, sets h=0, s=0
filter_hue [imagePath] [minValue] [maxValue] <-- shows only pixels with hue from given range (hue has values from [0, 360])
filter_skin [imagePath] <-- filter the color of human skin (using HSL color space)
filter_red_eyes [imagePath] <-- filter red eyes on photography

Noise filters:

average_rgb_filter [imagePath] <-- filter noises using moving average on RGB coordinates
median_rgb_filter [imagePath] <-- filter noises using moving median on RGB coordinates
average_y_filter [imagePath] <-- filter noises using moving average only on Y from YCbCr
median_y_filter [imagePath] <-- filter noises using moving median only on Y from YCbCr

Binarization:

binarize [imagePath] [threshold] <-- binarize image using threshold (from 0 to 256)
binarize_otsu [imagePath] <-- binarize image using Otsu's method for calculating treshold
binarize_bernsen [imagePath] <-- binarize image using Bernsen's method for calculating treshold
binarize_mixed [imagePath] [threshold] <-- binarize image using mixed Otsu's and Bernsen's method with a given threshold

Morphology:

Meanings of arguments in the following commands:
shape = type of structural element. Can be: "Square", "Circle", "Cross" (write without quotation marks)
windowSize = non-negative integer. 0 means just 1 element, 1 = 9 elements (3x3), 2 = (5x5) etc.
erosion [imagePath] [shape] [windowSize] <-- apply erosion (on otsu binarized image)
dilation [imagePath] [shape] [windowSize] <-- apply dilation (on otsu binarized image)
opening [imagePath] [shape] [windowSize] <-- Dilation after erosion (otsu binarized)
closing [imagePath] [shape] [windowSize] <-- Erosion after dilation (otsu binarized)
lut_erosion [imagePath] [shape] [windowSize] <-- Same as above, but using LUT (advice: do not use with windowSize > 1)
lut_dilation [imagePath] [shape] [windowSize]
lut_opening [imagePath] [shape] [windowSize]
lut_closing [imagePath] [shape] [windowSize]
rgb_erosion [imagePath] [shape] [windowSize] <-- apply erosion (on original RGB image)
rgb_dilation [imagePath] [shape] [windowSize] <-- apply dilation (on original RGB image)
convex_hull [imagePath] [maxIterations] <-- convex hull over binarized image
skeleton [imagePath] [maxIterations] <-- find skeleton of the object (over binarized image)
pruning [imagePath] [maxIterations] <-- pruning of the object (over binarized image)