抄録
A re-extraction scheme is presented for chromatically complex object in noisy scene. Image features of target objects are described in terms of color sampeles with a palette represented by normalized trichromatic vectors. Within the framework of Euclidean geometry, chromatic consistency is defined and applied to pixelwise matching of oberrved imagery with the palette. 2D expansion of the palette is evaluated for indexing intensity discrepancy due to the object imasges. The chromatic consistency and intensity discrepancy are jointly exploited to re-extract target objects.
