CAJ | 학술논문

Guo等人利用n个水平集方程构造n个区域提出一种改进的CV模型(简称 MCV模型),该模型需要的迭代次数很少,提高了图像分割的效率,但其分割结果受初始曲线位置的影响较大,极易陷入局部最优,无法分割复杂图像,且利用传统的Heviside函数无法得到准确的均值信息,因此无法保证数值的稳定性.本文对MCV模型进行改进,先对网像进行预分割得到初始曲线以提高分割效率且能保证分割结果全局最优,构造新的符号函数取代传统的Heviside函数改进MCV模型以保证数值稳定性.对MR图像进行的分割实验表明,其在保证迭代次数较少的同时分割更加准确.
Guo등인이용n개수평집방정구조n개구역제출일충개진적CV모형(간칭 MCV모형),해모형수요적질대차수흔소,제고료도상분할적효솔,단기분할결과수초시곡선위치적영향교대,겁역함입국부최우,무법분할복잡도상,차이용전통적Heviside함수무법득도준학적균치신식,인차무법보증수치적은정성.본문대MCV모형진행개진,선대망상진행예분할득도초시곡선이제고분할효솔차능보증분할결과전국최우,구조신적부호함수취대전통적Heviside함수개진MCV모형이보증수치은정성.대MR도상진행적분할실험표명,기재보증질대차수교소적동시분할경가준학.
Guo proposed an improved CV model (MCV model) that needs less iteration, but trapped in local optima for the influence of regions of initial contours; also some points were segmented in a wrong region or were omitted. The means gated by traditional Heviside function isn't accurate to keep the numerical stability. In this article we modify the MCV model, propose a new model using n equations of level set to structure n Regions: pre-segment the image to get the initial contours to avoid the results trapping in local optima and improve the efficiency of segmentation. Then we modify the MCV model by structuring a new symbol function to replace the Heviside function which can keep the numerical stability. Experiment results show that the new model can obtain good results efficiently.