中文摘要：

针对彩色图像分割中分水岭算法的过分割问题,提出了一种改进的基于标记提取的分水岭算法.改进后的算法由平滑滤波、彩色梯度计算、标记提取和分水岭变换组成.在平滑滤波阶段,设计了保边性能优于传统频域低通滤波器的频谱包络滤波器并运用于彩色图像及其梯度的平滑.彩色图像梯度计算直接在彩色向量空间进行.在标记提取阶段,利用局部极小值区的深度信息自适应控制扩展最小变换在平滑后的梯度图像中提取标记,然后融合极小值区的多重信息修改标记并将其叠加到原始梯度图像.对叠加标记后的梯度图像进行分水岭变换即得到最终的分割结果.实验结果表明,改进后的算法克服了传统算法边缘定位不准以及弱边缘提取困难等问题,参数选取更加合理,自适应程度提高.

英文摘要：

A modified marker-extraction based watershed algorithm was proposed in this paper to deal with the over-segmentation during color image segmentation.The modified algorithm was constituted of smooth filtering,color gradient calculation,marker-extraction and watershed transformation.During smooth filtering,a novel spectrum envelope filter was designed.The new filter had a better performance on edge-preserving which was used to smooth the imported color image and gradient image.The color gradient was calculated right in the color vector space.During the course of marker-extraction,H-minima transformation was used to extract minima-marker in smoothed gradient image firstly,whose parameter was adaptively controlled by the depth information of local minima region.Then,the extracted minima-marker was updated by more information of local minima region.Finally,the updated minima-marker was imposed on the original gradient image to get the marked gradient.The final result was gotten from the watershed transformation on the marked gradient.The experimental results indicate that the modified algorithm overcomes the difficulties in getting accurate edges and detecting weak edges.Furthermore,it has a more reasonable initialization rule of parameters and a better adaptability.