Abstract: To solve the problem of large calculation and sensitivity to noise disturbance in image segmentation, this paper combines a novel intelligence optimization algorithm, i.e. the artificial fish swarm algorithm, with  discrete wavelet transform, and proposes a parallel segmentation method for noise-polluted images. The suggested method employs a special scheme to decide the step of individuals in the fish swarm to improve the speed of convergence, which integrates the fixed step and the adaptive step. On the other hand, discrete wavelet transform is introduced to improve the signal-to-noise ratio of segmented images, in which appropriate wavelet coefficients are selected to reconstruct a noise-suppressed image. The 2D Otsu method serves as the fitness function for the improved AFSA to obtain an optimal threshold. Experimental results show that the proposed method is superior to the method proposed by Pan and Wu in terms of segmenting effect and noise reduction.

Key words: image segmentation, Artificial Fish Swarm algorithm（AFSA）, wavelet transform, noise reduction

摘要： 针对传统的图像分割方法计算量大、抗噪性弱等问题，将新型的智能仿生优化算法——人工鱼群算法（Artificial Fish Swarm Algorithm，AFSA）和小波变换有效地应用到图像分割中，并提出一种并行的阈值分割方法。采用合适的固定步长与自适应步长相结合的方法提高AFSA收敛速度，利用小波变换对小波系数进行阈值处理来提升图像信噪比。利用二维Otsu作为人工鱼群算法的适应度函数，以获得最优阈值。实验结果显示，该方法在分割质量和降噪方面较潘喆等人提出的方法有明显提高。

关键词: 图像分割, 人工鱼群算法, 小波变换, 噪声抑制