Developing an Automatic Method for Generating Colour Palettes from Landscape Images

In the design process, particularly in landscape, architecture and urban design, colour is a vital element. Designers often use a specific colour palette to clarify their ideas or colourise their plans. Landscape is one of the most significant inspirations for designers; in light of the specific colour matching in the landscape, designers can build a colour palette with the local to indicate the theme of their plan. For instance, a typical landscape view with red bricks, greenish moors and blue sky as well as the white cloud in the United Kingdom. This suggests that a colour palette derived from landscapes can represent particular characteristic features of a place, which can relate to the themes and emotions of the designers. Designers often work on colour palette generation based on subjective colour assessment.

However, this work focuses on not only the subjective but the objective method to extract colour palettes from landscape images. This work was mainly divided into four parts, colour-palette generation by designers (novice and professional designers), predicting visual colour differences between pairs of colour palettes, and colour palette extraction using either K-means clustering or an eye-tracking system. A psychophysical experiment in which 30 (design-based) participants each selected five colours from each of a set of images was first conducted to archive the designer extraction method and the palettes collected from this method were used as the ground-truth data against other (more automatic) methods in this research. The palette difference prediction metric was explored based on the second experiment in which a total of 95 pairs of palettes were rated for visual difference by 20 participants. This leaded to one algorithm (MICDM), which was used as the measure to compare the palette extraction using K-means clustering methods in RGB and CIELAB colour spaces. The eye-tracking experiment was undertaken with the same participants and image stimulus to obtain the eye-tracking data. The eye-tracking data was analysed and modified into colour palettes as the extraction method from the eye-tracking data.

Overall, the experiments and methods described in this work performed to generate colour palettes from landscape images. The palette extraction from eye tracking performs better than K-means clustering, and may have future application to design.