多尺度变换与标记符分水岭相结合的图像分割

Image segmentation on the combination of multi-scale transform and marker watershed

针对传统分水岭算法对噪声敏感和易于产生过分割的问题,提出一种将多尺度变换与控制标记符分水岭算法相结合的分割策略。首先对原始图像进行高斯金字塔尺度变换,抑制部分噪声,降低细节干扰;再用Sobel梯度算子计算尺度变换后的图像梯度;然后对梯度图像进行重建,再采用基于控制标记符的分水岭变换算法对重建后的梯度图进行分割,最后将分割结果变换回原始尺度。实验结果表明,该方法能够抑制传统算法中的过分割问题,且分割效果较好。

Aimed at resolving the problems of sensitivity to noise and over- segmentation existing in traditional watershed algorithm, an image segmentation strategy on the combination of multi - scale transform and marker controlled watershed is proposed. Firstly, in order to restrain part of the noise and reduce the interference of details, we use Gaussian pyramid to transform the scale of original image. Secondly, we employ the Sobel gradient operator to calculate the image gradient amplitude. Thirdly, we reconstruct it and apply marker controlled watershed algorithm to segment the reconstructed gradient image. Finally, we transform the segmentation results to original scale. Experimental results show that the strategy in this paper can restrain the over - segmentation of traditional algorithm, and has a good effect.