Abstract: Miao costume images have problems such as embroidery line texture, complex shape and large color difference. In view of these problems, an image segmentation algorithm of Miao costumes based on fuzzy fitting image is proposed. Firstly, the fuzzy local and global fitting images are defined in the image fuzzy region. Meanwhile, according to the nature of Kullback-Leibler divergence used to describe the difference between two probability distributions, the fuzzy energy function is constructed based on the image difference between the original image and the fuzzy local and global fitting images in the Kullback-Leibler divergence, which can drive the initial contour to the target boundary. Next, the adaptive weight is defined by using the normalized intra-class variances of the pixel grayscale inside and outside the contour curves of the image, which can automatically adjust the parameters between the local and global fuzzy energy terms. Finally, we add a regularization term and a length term in the energy function, and an edge detector is introduced into the regularization term and length term to smooth the edge of the image. In the experiment, the validity of the proposed algorithm is verified by experiments on natural images. The similarity and sensitivity coefficients of the segmentation results are above 0.978 and 0.981, respectively. Then, the proposed optimization algorithm is applied to the Miao costume image segmentation. The results show that the proposed algorithm has good segmentation results, and it requires a few iterations and a little segmentation time. Furthermore, the segmentation results are robust to the initial level set function and the Miao costume images.

Key words: Miao costume, image segmentation, active contour, fitting image, Kullback-Leibler divergence, adaptive weight

摘要： 针对苗族服饰图像存在绣线纹理、形状复杂及色彩差异性的问题,提出了一种基于模糊拟合图像驱动的苗族服饰图像分割算法。首先,结合局部与全局图像信息在模糊区域中定义了模糊局部与全局拟合图像。同时,根据Kullback-Leibler散度用于描述两个概率分布之间差异的性质,利用原始图像和拟合的模糊局部与全局图像在Kullback-Leibler散度方面的图像差异,构造了模糊能量函数,从而驱动初始轮廓曲线向目标边界移动。其次,采用局部与全局轮廓曲线内外区域的像素灰度归一化类内方差构造自适应权重系数,以此来自动调整全局与局部模糊能量项之间的参数。最后,在能量函数中添加了一个正则项与长度项,并在其中引入边缘检测器,来平滑图像边缘。该算法的性能通过在自然图像上进行验证,分割结果的相似性系数与敏感性系数分别达到0.978和0.981以上。然后使用提出的算法在苗族服饰图像上进行实验,结果表明提出的算法具有较好的分割结果,并只需要较少的迭代次数与分割时间。此外,分割结果对水平集函数的初始化与苗族服饰图像具有鲁棒性。

关键词: 苗族服饰, 图像分割, 主动轮廓, 拟合图像, Kullback-Leibler散度, 自适应权重