基于隐马尔科夫模型的激光主动成像图像分割方法  

Image segmentation method of laser active imaging based on hidden Markov model

作　　者：余楠 晋玉星[1] YU Nan;JIN Yuxing(Information Engineering Collage of Kaifeng University,Kaifeng Henan 475000,China)

机构地区：[1]开封大学信息工程学院,河南开封475000

出　　处：《激光杂志》2022年第2期105-109,共5页Laser Journal

基　　金：河南省重点研发与推广专项(科技攻关)(No.212102210106)。

摘　　要：为了提高分割结果一致性,更为详细地凸显激光图像特征,提出一种基于隐马尔科夫模型的激光主动成像图像分割方法。通过小波转换获得图像同一坐标的不同频带信息,同时依靠二维多分辨率分解划分噪声,构建尺度空间,依据Wiener滤波与高斯混合模型去除对图像内冗余去噪,将处理后图像储存在尺度空间内,使用小波域隐马尔科夫模型提取图像的边沿与局部特征,并将图像超像素信息融入模型距离函数与点对先验几率内,提升分割结果的质量,最后计算图像像素迭代强度、区域级迭代强度与像素对于所属超像素的贡献度,得到图像内局部区域的边沿,实现激光主动成像图像的分割。实验证明,所提方法的图像一致性结果平均值为95%,且具有分割边缘清晰的优点。In order to improve consistency of segmentation results and highlight characteristics of laser image in more details, a laser active imaging image segmentation method based on hidden Markov model is proposed. The different frequency band information of the same coordinate of image are obtained through wavelet transformation. At the same time, the scale space is constructed by noise dividing rely on two-dimensional multi-resolution decomposition. The redundant noise in image is removed according to the Wiener filter and the Gaussian mixture model. The processed image is stored in the scale space, and the wavelet domain hidden Markov model is used to extract the edge and local features of image. The image superpixel information is integrated into the model distance function and the prior probability of the point pair to improve quality of segmentation results. Finally, the iterative intensity of image pixels, the area-level iterative intensity and the contribution of pixel to superpixel are calculated. The edge of local area in image is obtained to realize segmentation of laser active imaging image. Experiments show that average image consistency result of proposed method is 95%, and it has the advantage of clear segmentation edges.

关 键 词：隐马尔科夫模型 激光主动成像 图像分割 点对先验几率 小波转换 

分 类 号：TN249[电子电信—物理电子学]