基于区域特征合并的分水岭图像分割

传统分水岭变换图像分割方法容易造成过度分割,不利于后期的图像分析与处理。采用一种分水岭变换结合区域特征合并的方法,首先将原始图像进行形态学变换,获取梯度图像;再进行分水岭变换,并将运算后的结果进行基于区域纹理、灰度一致性及区域平滑性等特征的区域合并;最终获取分割结果。实验结果表明,与传统分水岭变换方法相比,该方法能够有效降低因噪声、明暗纹理产生的过分割现象,方法可行、有效。

Watershed Image Segmentation Based on Region Feature Merging

Excessive segmentation is easily caused by traditional watershed transform image segmentation method and it is disadvantaged to image analysis and processing at the post stage. A method incorporating with watershed transform and region feature merging is adopted. Firstly, the original image is transformed morphologically to get gradient image. Then watershed transform is performed and the operated result is regional merged based on region texture, gray level consistency and region smoothness. And the segmentation result is got finally. Experimental results show that the method is feasible and effective comparing with traditional watershed transform method and excessive segmentation due to noise and light-to-dark texture is reduced effectively.