基于数学规划条件的粘连颗粒图像鞍点搜寻

Programming Restraint Based Saddle-point Identification for Granule Images

鞍点在图像粘连对象分割的过程中起着非常关键的作用,目前已有若干以分割为目的的鞍点搜索算法,但在其简捷性、鲁棒性等方面仍有改善的余地。提出一种基于数学规划条件的颗粒图像鞍点搜寻新方法。该方法首先根据颗粒图像3维地貌化后得到的像素空间分布信息,利用数学规划原理提出颗粒图像的鞍点规划条件,然后根据其条件进行鞍点检测。将提出的算法应用在真实粘连颗粒图像上,实验结果表明,该算法具有较高的准判率,为颗粒图像的分割以及相关后续处理奠定了理论分析基础。

Saddle point has been used for segmenting the connected granules in the image. To date there have been several reports on separating algorithms, but there are still space for the improvement of the robustness and simplicity. Based on mathematical programming conditions, the author proposes a new approach to search for the saddle point. First the original image is transformed to 3-D topographic counterpart, then combined with pixel information and saddle point restraints set by the programming premises, the desired saddle point can thus identified. The algorithm is put to test on given images, and the results show certain adequacies in robustness, simplicity, and correction rate.