kdanik / scala-kuva

Scala library for image and color processing, manipulation and filters

Description

Scala Kuva is a Scala library that focuses on image and color processing and filtering.

Functionality

Color types

The library contains implementations for three color formats: RGB(A), HSL(A), and HSV(A). These color types are immutable and can be converted between each other. Additionally, they can also be converted to java.awt.Color and created from it

Immutable image

For all image operations in this library own implementation of the immutable image is used. The Immutable image simply wraps two-dimensional Vector of immutable Pixels, offering some useful operations to work with it in functional way.

It can also be converted to java.awt.image.BufferedImage and created from it.

Color and image manipulation

This library contains implementation for various color and image operations. For example: most grayscale algorithms, many blend modes, binarization, resizing with different algorithms, croping and more.

This initial image will be used to demonstrate the effects of the different operations.

Basic image operations

Cropping

Scaling

Nearest Neighbor

Example image is downscaled (from 1024x1280 to 400 x 500) using nearest neighbor algorithm.

Example image is upscaled (from 128x160 to 1024x1280) using nearest neighbor algorithm.

Bilinear interpolation

Example image is upscaled (from 128x160 to 1024x1280) using bilinear interpolation.

Bicubic interpolation

Example image is upscaled (from 128x160 to 1024x1280) using bicubic interpolation.

Hashing

Average hash (aHash)

The average hash algorithm works by converting an image into a grayscale, downsampling it to a small size (8x8 per default), computing the average pixel value, and creating a binary hash by comparing each pixels value with the average value.

Median hash (mHash)

The average hash algorithm works by converting an image into a grayscale, downsampling it to a small size (8x8 per default), computing the median pixel value, and creating a binary hash by comparing each pixels value with the median value.

Difference hash (dHash)

The difference hash algorithm works by converting an image into a grayscale, downsampling it to a small size (9x8 per default), and by computing the difference in pixel values between adjacent pixels (compares each pixel with the pixel on the right from it).

Blur

Box blur

Median blur

Gaussian blur

Grayscale

Averaging

This algorithm simply uses average of all color channels as grayscale color.

Decomposition (using max channel value)

Calculates max value of 3 color channels and uses it as grayscale color.

Decomposition (using min channel value)

Calculates min value of 3 color channels and uses it as grayscale color.

Desaturation

This algorithm first convert color to HSL and then reduces saturation to 0.

Luma (ITU-R BT 601)

Calculate grayscale color using luma (luminance or brightness) formula (from ITU-R BT 601 specification).

Luma (ITU-R BT 709)

Also calculates grayscale color using luma formula, but from newer standard (ITU-R BT 709). The weighting coefficients used in the formulas are different between the two standards.

Single channel (red)

Uses as grayscale color the value of red color channel

Single channel (green)

Uses as grayscale color the value of green color channel

Single channel (blue)

Uses as grayscale color the value of blue color channel

Lightness

Calculates and uses as grayscale lightness using formula from this paper.

Lightness (HSL)

Calculates and uses as grayscale lightness using formula from HSL color format.

Middle value

Calculates and uses as grayscale middle value of the RGB channels

N-shades of gray (using Luma 601 for initial grayscale)

Converts grayscale image to grayscale image with limited number of shades. (Example image has 8 shades of gray)

Inversion

Color inversion is a photo effect that flips all colors to their opposite hue on the color wheel (with inversion of alpha channel or without).

Blending colors

To demonstrate different color blending algorithms two abstract (background and foreground) images will be used. (There also additional examples in src/main/resources/source/blend/gradient folder)

Background (initial image):

Foreground (overlay image):

Simple Alpha Compositing

Dissolve

Multiply

Screen

Overlay

Hard Light

Soft Light

Color Dodge

Linear Dodge

Color Burn

Linear Burn

Vivid Light

Linear Light

Divide

Difference

Subtract

Lighten only

Darken only

Hard mix

Pin light

Reflect

Exclusion

Geometric mean

Luminosity

Color

Binarization

Binarization is the process of converting a multi-tone grayscale image into a two-tone black and white image. This is achieved by finding a threshold value, which is then used to create the binary output.

Otsu's method

Otsu's method is a thresholding technique used to separate an image into foreground and background pixels. It calculates the optimal threshold value by maximizing the between-class variance of pixel intensities.

Otsu's method requires the image to be converted into grayscale before applying the algorithm, and the resulting threshold value may vary depending on the specific grayscale conversion algorithm used.