基于Contourlet变换的图像智能分割方法研究  

Research on Image Segmentation Based on Contourlet Transform

作　　者：刘海[1] 

机构地区：[1]广东科学技术职业学院软件学院,广东珠海519090

出　　处：《计算机测量与控制》2018年第1期202-205,共4页Computer Measurement &Control

基　　金：广东省高职教育品牌专业(2016gzpp007)

摘　　要：由于获取图像的设备不同,不同设备所拍摄图像的质量、空间分布特性差异较大;采用当前图像分割方法对图像噪声进行分割时,容易产生过分割和奇异扩散现象;为此,提出一种基于Contourlet变换的图像智能分割方法;该方法先采用方向滤波器组实现图像各个方向纹理分离,利用小波变换替换拉普拉斯变换进行图像多个方向子带分解,计算图像多尺度似然函数,并依据图像最大似然函数估计准则获得图像的初始分割,在此基础上采用自适应上下文结构从图像粗尺度的分割结果融合至最细尺度,获得最终的图像智能分割结果;仿真实验结果表明,所提方法能够有效消除噪声对图像分割的影响,使图像分割精度更高,且运行时间更短,该方法在图像获取设备中具有较高的实践价值。Due to the different images of the equipment,the quality and spatial distribution of the images taken by different devices are quite different.The current image segmentation method is sensitive to noise,and the segmented image results are prone to over-segmentation and singular diffusion.In this paper,an image segmentation method based on Contourlet transform is proposed.The method uses the direction filter bank to realize the texture separation in each direction of the image,and substitutes the Laplace transform with the wavelet transform to carry on the multi-scale sub-band decomposition of the image,calculates the image multi-scale likelihood function,and according to the image maximum likelihood function estimation criterion the initial segmentation of the image is obtained.On this basis,the adaptive context structure is used to fuse from the roughness of the image to the finest dimension to obtain the final image segmentation result.The experimental results show that the proposed method can effectively eliminate the influence of noise on image segmentation and achieves high image segmentation accuracy.

关 键 词：CONTOURLET变换 图像智能 智能分割方法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]