针对现有寻优算法存在的重复计算问题, 提出了基于递推遗传的模糊3-划分熵多阈值荧光原位杂交(Fluorescence in Situ Hybridization, FISH)基因提取算法来提高用模糊划分熵算法提取多阈值FISH基因的效率。采用迭代验证法确定隶属度函数窗宽, 并使用附加边界条件及灰度权重的隶属度函数对图像进行模糊3-划分。为了提高阈值寻优的效率, 引入递推算法将模糊熵的计算转化为递推过程, 并保存部分不重复的递推结果用于后续的计算, 最后采用遗传算法寻优, 使得种群个体的计算能使用预存结果快速搜索全局最优阈值。对提取结果与几种常用算法进行了直观比较, 并对处理时间、分类概率等性能指标进行了量化分析。对多幅不同类型的仿真人工图像和真实FISH图像的测试表明, 处理时间仅为常用算法的1%, 错误划分概率小于6.00×10-2。提出的算法可以准确, 高效地提取FISH基因目标。

A new fuzzy 3-partition entropy approach based on a fast recursive genetic algorithm was proposed to reduce the repeated computations and to improve the processing efficiency in extraction of FISH-labelled (Fluorescence In Situ Hybridization) genes. An iteration validation method was presented to determine the window width of the membership functions and the membership functions considering the boundary conditions and gray weights were selected to perform the fuzzy 3-partition. To improve the efficiency of selecting optimal thresholds, a recursive algorithm was presented to convert the computation of fuzzy entropy to a recursive process. Then, the no-repetitive results of the processing moments were stored for the succeeding genetic algorithm to compute the fitness of each individual. Finally, the optimal thresholds were searched by the genetic algorithm in a high speed. The result of the proposed algorithm was compared to those of the several common algorithms and the classification probability and run time were analyzed as the test criterion of optimal thresholds. By evaluating various types of simulated images and real FISH images, it shows that the run time of the proposed algorithm is 1% that of other common algorithms and the misclassification error is less than 6.00×10-2. These results demonstrate that the proposed algorithm is effective for improving the precision and efficiency of extracting FISH-labelled genes.