Abstract:

Multi-threshold image segmentation is a simple, accurate, efficient, and universal, image segmentation method. Compared with single-threshold image segmentation, it is more suitable for color images containing much information. In multi-threshold image segmentation, the amount of calculation with the traditional enumeration method surges with the increase of threshold number. This takes more time to segment a color image and reduces the precision of segmentation.Therefore, a multi-threshold image segmentation method based on the Improved Northern Goshawk Optimization(INGO) algorithm is proposed. Cubic chaos optimization and lens imaging reverse learning strategy are used to increase population diversity while optimizing initial solutions, which expands the scope of population search, enables the INGO algorithm to better search for potential optimal solutions, and enhances the search ability of the algorithm.A hybrid strategy that combines the optimal worst case inversion with a lens imaging inverse learning strategy can help the INGO algorithm avoid situations where it is prone to fall into a local optimum and improve convergence accuracy.The experimental results of multi-threshold color image segmentation on classic Berkeley test images show that among the algorithms, such as GWO, PSO, and ChOA, the proportion of INGO algorithm achieving the optimal average of Peak Signal-to-Noise Ratio(PSNR) and Feature Similarity Index Mersure(FSIM) is 100.000% and 78.125%, respectively. It achieves better image segmentation results while ensuring the convergence efficiency of the algorithm and has strong theoretical application value in the field of multi-threshold image segmentation.

Key words: Northern Goshawk Optimization (NGO), multi-threshold segmentation, symmetric cross-entropy, cubic chaos, lens imaging reverse learning strategy

摘要：

多阈值图像分割是一种简单、准确、高效且普遍的图像分割方法，相比单阈值图像分割更适用于包含大量信息的彩色图像。在多阈值图像分割中，随着阈值数量的增加，传统的枚举法计算量增大，分割一幅彩色图像不仅需要更多的时间，而且分割精度也随之降低。提出一种基于改进北方苍鹰优化（INGO）算法的多阈值图像分割方法。利用立方混沌优化与透镜成像反向学习策略增加种群多样性，在优化初始解的同时扩大种群搜索范围，使INGO算法尽可能搜索到潜在的最优解，增强算法的搜索能力。将最优最差反向与透镜成像反向学习策略相结合，避免INGO算法易陷入局部最优的情况，提高收敛精度。在对经典的伯克利测试图像进行多阈值彩色图像分割的实验结果表明，在GWO、PSO、ChOA等算法中，INGO算法取得峰值信噪比和特征相似度最优平均值的占比分别为100.000%和78.125%，在保证算法收敛效率的同时获得较优的图像分割结果，在多阈值图像分割领域具有较强的理论应用价值。

关键词: 北方苍鹰优化, 多阈值分割, 对称交叉熵, 立方混沌, 透镜成像反向学习策略