基于Unit-Linking PCNN和图像熵的图像分割新方法

Image Segmentation New Methods Using Unit-Linking PCNN and Image’s Entropy

基于单位链接脉冲耦合神经网络(Unit-Linking PCNN)模型,在图像方差准则基础上将最大香农熵准则,最小交叉熵准则相结合,再组合图像分块策略与图像加权预处理策略,提出了不同准则和策略组合的8种图像分割方案。分析了各种准则和策略的优势和不足,比较了各种分割方案条件下的图像分割效果,部分方案的分割结果体现了图像的更多的细节。与已有文献结果比较,具有使用神经元模型参数少的特点,与PCNN模型相比,该模型参数对图像分割结果的影响较不敏感。计算机仿真结果表明,该方法具有较好的图像分割效果和实验仿真速度性能,具有较强适用性。

Based on the model of Unit-Linking PCNN, maximum Shannon entropy rule, minimum cross-entropy rule and some image pre-processing strategy were introduced for suggesting eight image segmentation schemes under different rules and strategies. Then the advantages diversified rules and strategies were analyzed and compared. Compared with existent correlative method, which have less parameters of the neural networks model, the parameters of models are less sensitive to image segmentation results than PCNN model. Simulation results show that the methods have both preferable segmentation results, fast speed and fine adaptability.