摘要
在基于边缘的轮廓提取算法中,边缘检测精度直接影响轮廓提取的效果,而边缘检测算法通常分为图像滤波和边缘提取两个步骤。图像滤波算法是以损失图像真实信息为代价达到滤除噪声的效果(通常称为图像失真),图像失真的程度直接影响后续边缘提取的准确度;边缘提取算法进行无区分性的全局图像梯度检测,然而以轮廓提取为目标的边缘检测算法期望避免检测到图像中过于琐碎的细节,以减少轮廓提取的干扰信息。针对上述问题,提出了一种辅助轮廓提取的边缘检测算法,即结合RPCA滤波和自适应Mean-shift聚类的图像边缘检测算法。首先将RPCA算法引入到图像滤波,同时定义图像保真的度量指标,优化影响RPCA算法图像估计准确度的权重因子;然后将均值-标准差之比引入Mean-shift算法,使其能够自适应识别变化平缓区域和变化剧烈区域;最后应用Canny算子对图像进行边缘提取。实验结果表明,RPCA滤波算法能实现对原图像信息的最优估计,自适应Mean-shift算法能有效抑制图像中过于琐碎的细节且保留边缘信息。该算法能准确提取到图像中主要边缘信息而忽略图像细节信息,且适用于不同类型的图像。
In the edge-based contour extraction algorithm,the edge detection accuracy directly affects the contour extraction effect,while the edge detection algorithm is usually divided into two steps:image filtering and edge extraction.The image filtering algorithm achieves the effect of filtering out noise at the expense of the true information of the image(usually called image distortion).The degree of image distortion directly affects the accuracy of subsequent edge extraction.The edge extraction algorithm performs non-discriminant global image gradient detection.However,the edge detection algorithm aiming at contour extraction is expected to avoid detecting too trivial details in the image to reduce the interference information extracted by the contour.Aiming at the above problems,an edge detection algorithm for auxiliary contour extraction is proposed,which combines RPCA filtering and adaptive Mean-shift clustering.Firstly,the RPCA algorithm is introduced into image filtering,and the image fidelity metric is defined to optimize the weighting factor that affects the image estimation accuracy of the RPCA algorithm.Then the ratio of mean-standard deviation is introduced into the Mean-shift algorithm,which makes it possible to adaptively identify the gradual change region and the violent region.Finally,the Canny operator is used to extract the edge of the image.The experiment proves that the RPCA filtering algorithm can achieve the optimal estimation of the original image information.The Mean-shift algorithm can effectively suppress the trivial details in the image and preserve the edge information.The proposed algorithm can accurately extract the main edge information in the image and ignore the image detail information,which is suitable for different types of images.
作者
刘文振
傅惠南
何金彬
LIU Wen-zhen;FU Hui-nan;HE Jin-bin(School of Electro-mechanical Engineering,Guangdong University of Technology,Guangzhou 510006,China)
出处
《计算机技术与发展》
2019年第11期27-31,共5页
Computer Technology and Development
基金
国家自然科学基金(50675037)
