Image Colourisation

Image

Select image

Upload / Take Photo

Scale

Algorithm & Parameters

POINT SAMPLING

Select method

Width Height

n sample points

Seed

Random Seed

Point Sampling

N_fraction pixels sampled

Fill_fraction spatial-fill share

ξ edge / smooth density ratio

Seed point sampling

Random Seed

n = —

Parameters

Select Kernel

σ₁ spatial bandwidth

σ₂ intensity bandwidth

p intensity exponent

δ regularisation parameter

Parameters

β₁ σ₁ coverage

β₂ σ₂ coverage

σ₁ = — · σ₂ = —

RKHS colourisation reconstructs the full colour image from a small set of known sample points using kernel interpolation.

(a) Scaled image

Show pipeline stages

(b) Greyscale

(c) Raw edge mask

(d) Smoothed edge mask

(b) Greyscale with data points D

(c) Colourised output

‖·‖_F — smaller is better

SSIM — closer to 1

Results

Method	Frobenius	SSIM

‖·‖_F — smaller is better

SSIM — closer to 1 is better