增强图像细节和去噪能力的改进形态学分水岭算法

为了抑制分水岭算法过分割和滤波后保持图像细节，论文提出一种改进的 形态学分水岭分割算法。首先，对图像进行多尺度小波分解得到低频系数和高频系数；对低 频系数进行基于Perona-Malik 扩散模型各向异性扩散滤波；对高频系数，引入神经网络中的 sigmoid 函数改进自适应遗传算法的变异和交叉概率生成，并用父代的最优个体替换子代中 最差的个体来保护最优个体不被破坏，克服遗传算法的局部最优现象，利用改进的自适应遗 传算法增强和去噪。然后，对梯度图像做锐化处理以突出边缘, 再做形态学运算并进行 H-minima 标记。最后，执行分水岭分割，实现改进的算法。实验结果表明，改进算法能够 有效地抑制噪声的干扰，减轻过分割，分割精度也有所提高。

Improved Morphological Watershed Algorithm toEnhance Image Detail and Denoise Ability

An improved morphological watershed algorithm is presented to solve the over-segmentation of watershed and keep the image detail. First, the image is decomposed by multi-scale wavelet into high-frequency and low-frequency coefficients. The low frequency coefficient of the image is filtered by Anisotropic diffusion filter algorithm based on Perona-Malik diffusion model. The high frequency coefficient of the image is enhanced and denoised by using the improved adaptive Genetic Algorithm, which is gotten by introducing the sigmoid function in neural network to modify the generation of the mutation and crossover probability, and by replacing the worst individual of the children with the best individual of the father generation to protect the best individual from being destroyed, and to overcome the local optimal phenomenon of genetic algorithm. Then, the gradient image is sharpened to highlight the edge, the watershed algorithm is applied after conducting the morphological operation and the H-minima label on it, the improved algorithm is realized. Experiment shows that the improved algorithm can significantly restrain the noise disturbance and reduce the over-segmentation, and its segmentation accuracy is also improved.