自适应Mean Shift算法的彩色图像平滑与分割算法

采用Mean shift算法对图像进行平滑和分割处理时, 带宽和采样点权重的选择直接影响平滑和分割的效果. 带宽分为空域带宽和值域带宽. 本文根据图像颜色分布的丰富程度定义了自适应空域带宽. 在此基础上, 通过最小化局部方差函数和最大化频域结构相似度函数获得自适应值域带宽. 此外, 通过定义采样点权重, 克服了图像过平滑问题. 通过随机选取大量的图像进行实验, 结果表明运用本文所选择的带宽和权重, 可以得到正确的图像区域分割结果.

Adaptive Mean Shift Based Image Smoothing and Segmentation

Bandwidths and weights of sampling points are two key points in mean shift based image smoothing and segmentation. Bandwidths indicate spatial bandwidths and range bandwidths. Adaptive spatial bandwidths are defined according to color distribution of the image. Then, adaptive range bandwidths are obtained by minimizing the local variance function and maximizing the frequency structural similarity function. Additionally, weights of sampling points are defined to overcome over smoothness. Experimental results prove that the correct segmented regions are obtained by using the proposed bandwidths and weights.