迷彩设计中背景图像聚类方法的比较分析

迷彩设计中，通过对目标背景的图像信息进行归类处理，利用图像分割技术获取背景斑点，然后设计出与背景亮度和纹理相协调的迷彩图案. 目前，最常用的图像分割方法是均值聚类法，由于它直接利用灰度信息以致分割的细节不明显，容易导致模拟背景纹理失真. 针对它的不足，提出了采用基于自组织特征映射(self organizing feature map, SOFM)神经网络的分割方法对背景进行聚类分割. 该方法将整幅图像作为神经网络的输入，聚类分割后的颜色作为输 出，按照自组织特征映射网络的学习过程，使用其函数进行迭代运算直至学习停止. 比较实验结果表明，SOFM方法能更好地保留图像的细节纹理，得到较为理想的设计效果.

Comparison and Analysis for Image Clustering Methods of Camouflage Design

In the process of camouflage design, information of background images is classified, and background spots acquired using image segmentation. Camouflage patterns corresponding to luminance and natural textures of the background are designed. The common method of mean clustering uses the gray information for image segmentation, resulting in fuzzy details of textures. This paper proposes a new method based on self organizing feature map (SOFM) neural-networks for clustering segmentation. The method uses the entire image as input to the neural-networks, and colors after segmentation as output. Calculation is done iteratively using its special functions in the learning process of self-organizing feature map neural-networks until the learning stops. Experiments show that the SOFM method can preserve more texture details than the mean-clustering method and provide good results.