Color Image Classification Through Fitting of Implicit Surfaces

Accurate Color Classification and Segmentation for Mobile Robots

“…This is because shape of clouds of samples is affected by light incidence and reflection. Hence, a better technique of classification was proposed in (Alvarez et al, 2004) to overcome some of these drawbacks. This new approach is based on a technique used for 3D object reconstruction (Lim et al, 1995), and the use of implicit surfaces as the threshold that bounds and defines color classes.…”

“…A possible solution would include extracting the samples on the surface, but discarding samples within the surface would eliminate potentially valuable information for the process. Instead, (Alvarez et al, 2004) propose a different approach using the k-means algorithm. The k-means algorithm, as defined by Bishop (Bishop, 1996), divide a set of n point samples (x 1 , x 2 , … x n ) in a set of k disjoint, non-hierarchic sets (Q 1 … Q k ), through the minimization of a distance criterion d. Usually, Euclidian distance metric is used, which produces spherical clusters that, in this case, suit well as primitives for an implicit function.…”