Abstract: Given proper values of shift windows in spatial domain and color domain respectively, the color data were filtered by means of Mean-Shift clustering, during which the color data in the rcircular neighboring domain of current data point were used as the clustering samples. Then image data at current position were updated with the cluster center newly obtained. This algorithm overcomes the difficulty to choose proper window radius for the model of Mean-Shift filtering combining spatial domain and color domain to adopt the possible variation of image size. Finally, the experimental results verify the validity of MeanShift filtering.

Key words: color image filtering, Mean-Shift algorithm, spatial domain, color domain, window radius

摘要： 在空间域与色彩域中分别给定移动窗口半径的适当值，把空间位置处于当前数据点的圆邻域内的色彩数据作为算法的样本数据，利用窗口半径固定且独立于空间位置的核函数对色彩数据进行均值偏移(Mean-Shift)聚类，用聚类中心更新当前位置的图像数据，克服了空间域与色彩域结合的Mean-Shift图像滤波模型窗口半径难于恰当选取以适应图像尺寸变化的困难。实验证明了该算法的有效性。

关键词: 彩色图像滤波, Mean-Shift算法, 空间域, 色彩域, 窗口半径