Open AccessProbabilistic Gaussian similarity‐based local colour transfer
Author(s) -
Heo Y.S.,
Jung H.Y.
Publication year - 2016
Publication title -
electronics letters
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.375
H-Index - 146
eISSN - 1350-911X
pISSN - 0013-5194
DOI - 10.1049/el.2016.0632
Subject(s) - artificial intelligence , pattern recognition (psychology) , probabilistic logic , similarity (geometry) , gaussian , matching (statistics) , mathematics , mixture model , computer science , similarity measure , measure (data warehouse) , computer vision , image (mathematics) , statistics , data mining , physics , quantum mechanics
A simple and robust probabilistic colour transfer method using feature matching and Gaussian mixture model (GMM)‐based soft segmentation is proposed. Traditional colour transfer methods using GMM colour model require proper Gaussian matching where the euclidean distances of mean colours and positions of spatially overlapping regions are usually used as a matching measure. It is, however, difficult for these methods to find proper Gaussian matches for images taken from different illuminations, brightnesses, view‐points as well as different camera types and settings. This is troublesome since incorrect matches result in unwanted colour artefacts. To cope with this problem, a new Gaussian similarity measure and probabilistic colour transfer formulation are presented. Experimental results demonstrate that, compared with previous methods, more robust and proper colour transfer results are generated for images taken from different conditions.