Abstract: As it is of deficiency that the conventional Otsu algorithm is only applicable to single-threshold，the algorithm is extended to multi-threshold color image segmentation.Firstly，the selection of meaningful peak values among a group of maximum values is carried out.Then，according to the peak values，the diagram is divided into a certain number of segmentation intervals，in which the threshold is selected.Based on methodology of morphology，the segmentation is optimized.And the influence，as a result of ignoring the characteristics of space，which the threshold technique imposes on noises，is lessened.The results of the experiment indicate that the Otsu algorithm based on multi-threshold color image segmentation is able to implement automatic and rapid multi-threshold segmentation，and is noise-resistant to some extent.

Key words: color quantization, multi-threshold, the meaning peak, morphology

摘要： 针对传统Otsu算法只用于单阈值分割的不足，将Otsu算法推广到多阈值彩色图像分割中，提出先在众多极大值中寻找有意义峰值，根据峰值将直方图划分成多个待分割区间，再在每个区间进行阈值选取的方法；并且综合运用了形态学的方法对分割结果进行优化，降低阈值法因不考虑图像空间特性而造成的对噪声敏感的影响。实验结果表明，该方法能自动而快速地对彩色图像进行多阈值分割，而且具有较强的抗噪能力。

关键词: 颜色量化, 多阈值, 有意义峰值, 形态学