Robust Semantic Template Matching Using A Superpixel Region Binary Descriptor
Robust Semantic Template Matching Using A Superpixel Region Binary Descriptor
复制标题
使用超像素区域二进制描述符的鲁棒语义模板匹配
DOI：
10.1109/tip.2019.2893743
复制
发表时间：
2019
影响因子：
10.6
通讯作者：
Zhouping Yin
中科院分区：
文献类型：
--
作者：
Hua Yang;Chenghui Huang;Feiyue Wang;Kaiyou Song;Zhouping Yin
几乎所有的传统模板匹配方法都采用低层次的图像特征来度量模板图像和场景图像之间的相似性，使用相似性度量，如像素强度和像素梯度。虽然这些方法已被广泛用于许多应用中，但它们不能同时解决所有类型的鲁棒性挑战。在本文中，同时解决各种挑战的目标，我们提出了arobustsemantictemplate匹配（RSTM）的方法。受局部二进制描述符的启发，我们提出了一种新的超像素区域二进制描述符（SRBD）来构造RSTM的多级语义融合特征向量。SRB-Based是一种新的基于核距离的简单线性迭代聚类方法，用于从模板图像中提取稳定的像素点。然后，根据每个超像素区域与其相邻区域的平均亮度差，得到每个超像素的主导梯度方向，将每个超像素的特征描述为主导方向差向量，并将其编码为旋转不变的SRBD。在离线匹配阶段，RSTM的融合语义特征向量将多级SRBD特征与不同数量的超像素相结合。在在线匹配阶段，为了科普旋转不变性问题，提出了一种边缘概率模型，并将其应用于场景图像中模板图像的位置定位。此外，为了加速计算，采用图像金字塔。我们在从MS COCO数据集中随机选取的一个大数据集上进行了一系列的实验，充分分析了该方法的鲁棒性。实验结果表明，RSTM同时解决旋转变化，尺度变化，噪声，遮挡，模糊，非线性照明变化和变形具有高的时间效率，同时也优于以前的最先进的模板匹配方法。
Almost all conventionaltemplate-matchingmethods employ low-level image features to measure the similarity between atemplateimage and a scene imageusingsimilarity measures, such as pixel intensity and pixel gradient. Although these methods have been widelyusedin many applications, they cannot simultaneously address all types ofrobustnesschallenges. In this paper, with the goal of simultaneously addressing the various challenges, we present arobustsemantictemplate-matching(RSTM) approach. Inspired by the localbinarydescriptor, we propose a novelsuperpixelregionbinarydescriptor(SRBD) to construct a multilevelsemanticfusion feature vector for RSTM. SRBDusesa new kernel-distance-based simple linear iterative clustering method to extract the stablesuperpixelsfrom thetemplateimage. Then, based on the average intensity difference between eachsuperpixelregionand its neighbors, the dominant gradient orientation of eachsuperpixelcan be obtained, and thesemanticfeatures of eachsuperpixelcan be described as the dominant orientation difference vector, which is coded as the rotation-invariant SRBD. In the offlinematchingphase, the fusionsemanticfeature vector of RSTM combines the multilevel SRBD features with different numbers ofsuperpixels. In the onlinematchingphase, to cope with rotation invariance, a marginal probability model is proposed and applied to locate the positions oftemplateimages in the scene image. Moreover, to accelerate computation, an image pyramid is employed. We conduct a series of experiments on a large dataset randomly selected from the MS COCO dataset to fully analyze therobustnessof this approach. The experimental results show that RSTM simultaneously addresses rotation changes, scale changes, noise, occlusions, blur, nonlinear illumination changes, and deformation with high time efficiency while also outperforming the previous state-of-the-arttemplate-matchingmethods..