Sandra Skaff, Luca Marchesotti, Gabriela Csurka, Craig Saunders
CIC Color Imaging Conference, San Jose, California, November 7-11, 2011.
There has been extensive research on finding distance measures
between individual colors which conform to the human visual
system color perception. With the recent advent of using
and naming color combinations by abstract concepts (classic, romantic),
this paper addresses the new problem of computing distances
between such combinations, which are referred to as color
schemes. In addition, the paper proposes an algorithm to compute
the distance measure which is shown to be competitive compared
to various state of the art distance measures adapted to the
problem at hand. In particular, the proposed distance measure,
referred to as the Color-based Earth Moverâ??s Distance (CEMD)
embeds the CIEDE2000 color difference formula into the Earth
Moverâ??s Distance (EMD). The CEMD performance in computing
distances is evaluated through a color scheme retrieval framework.
Quantitatively, it is shown that the CEMD provides in general
the highest precision at K as compared with the EMD and a
distance based on Fisher Vector representations of color schemes,
which we refer to as FD. Qualitatively, it is shown that retrieved
color schemes are more similar to the query scheme when the
CEMD is employed as compared with the EMD and FD. Qualitative
results on image ranking by concept using the CEMD are
shown to be better than those obtained using tags.
Report number: