Colour Based Image Retrieval with Embedded Chromatic Contrast.
Xiaohong W. Gao,Yu Qian,Yuanlei Wang,Anthony White
DOI: https://doi.org/10.2352/cgiv.2012.6.1.art00026
2012-01-01
Abstract:Due to the over-whelming amount of digital images available in the internet, content-based image retrieval (CBIR) has been developed to complement with the current text-based approach. As such, colour has played a key role in representing image features and has been employed widely in such a development. However a colour appears differently to human eyes when it is viewed against different coloured backgrounds and surroundings, whereas none of existing colour spaces and models has taken this effect of colour contrast into account, leading to a number of unsatisfied retrieved results to a certain extent. This study aims to develop a colour appearance model/space to predict simultaneous colour contrast, which is in turn to be suitable on course to retrieve a collection of museum wallpaper papers. In doing so, a 2-field paradigm is maintained instead of traditionally 3- field one in an effort to model chromatic contrast, which has led to the extension of CIECAM02 into CIECAMcc. Colour based image retrieval is subsequently evaluated using 4 popular colour models and spaces, including CIECAMcc, CIECAM02, HSI, and RGB. Although it is unlikely to judge which method performs better purely based on colour content due to the nature of subjectivity in interpreting images, it can be said that in terms of both brightness and colourfulness contrast between foreground and background, CIECAMcc outperforms the others. In addition, CIECAMcc exhibits potentials in retrieving back images that constitute two shaded patterns the similar way as those depicted in a query image. However this phenomenon can not be simply explained away. Further investigation will be carried out in this regard in the future by including larger collections.