Color Quantization: Objective and Subjective Comparisons of Different Algorithms

摘要

This paper deals with quality assessment of color images dedicated to quantization algorithms. Two methods areuddescribed in details, one using the baker's transform and the other using the matrix of local pallets. In this assessmentudcampaign, the results of the described techniques are compared to those of standard algorithms such as median cut,udoctree and split & merge. Both objective and subjective assessment are performed. The need of subjective evaluationudcomes from the fact that the usual metrics do not integrate the HVS (Human Visual System). While the color is moreudconsidered as perceptual property than a quantitative data, these standard metrics fail in describing image distortion. Theudpreliminary results of the assessment campaign show that the described methods give good results with regards to theudformer ones.