基于形态学梯度重建的分水岭算法改进研究

Study on the Improvement of Watershed Algorithm Based on Morphological Gradient Reconstruction

图像分割是图像分析和计算机视觉领域中一个最重要的处理过程,常用的边缘检测算子存在边缘定位不准等缺点,而传统的分水岭算法则对噪声敏感且存在过分割的问题。本研究提出基于形态学梯度重建的改进型分水岭算法,该算法通过在人脸图像预处理过程中对图像进行中值滤波消除部分噪声,然后再对滤波后的图像进行增强对比度以及重构梯度图处理,最后利用分水岭算法对处理后的图像进行分割。仿真实验应用不同的算法对同一图像进行计算分割,对比结果显示,本研究提出的方法能够较好地抑制过分割,并且能够分割出人脸图像的主要特征。

Image segmentation is an important process in image analysis and computer vision field.In this study,an improved watershed algorithm based on morphological gradient reconstruction was proposed to deal with the problems of inaccurate orientation of common edge detection operator and over-segmentation of traditional watershed algorithm because of sensitiving to the noise.Combining with the current progress,the median filtering was adopted to eliminate the noise in the preprocessing of human face image.After filtering the image,the grey level transformation was used to increase contrast.At last the gradient map of the image was reconstructed and then the image was segmented by watershed algorithm.Comparing with the common edge detection operator and traditional watershed algorithm,the experimental results showed that this new algorithm can not only suppress the over-segmentation phenomena but also effectively segment the main features of the face.