目的 基于特征的匹配问题一直是计算机视觉和模式识别等领域中一项重要的基础研究。目前在刚性物体的匹配问题上，已经取得了较好的研究成果。然而在实际应用中，往往存在许多非刚性物体的匹配问题，需要对图像中存在的非刚性形变目标进行快速精确的配准，进而实现对图像的后续处理和分析。因此，实现快速而准确的非刚体匹配显得尤为重要。方法 针对传统特征点匹配方法在非刚性物体匹配中准确性差的问题，本文提出了一种基于DAISY算子和有约束Patch-Match的非刚体密集匹配算法。首先对参考图像和待匹配图像生成DAISY特征描述子，其次对两幅图像进行超像素分割，形成相互邻接但没有重叠的超像素块结构，并以其为单元，计算初始位置上对应每一个像素的DAISY特征算子聚合代价。然后，采用Patch-Match算法对整幅图像进行传播和变异，在变异过程中，通过图像预处理和分析得到的先验知识对位置标签的变异窗口进行局部空间约束，使得每个像素的位置标签在该空间范围内随机更新，计算新的聚合代价，保留代价较小的位置标签，重复迭代此过程，直到聚合代价不发生变化或者达到最大迭代次数为止。结果 实验选取了标准数据集、10幅分别由TFDS线阵列相机和框幅式相机采集的包含非刚体的图像进行匹配，均取得了较好的匹配效果，经验证，本文方法的匹配精度为86%，误匹配点的平均匹配误差为5个像素左右，误差是传统基于SIFT特征光流匹配方法误差的一半，并且本文采用的DAISY算子在特征提取速度上是Dense SIFT特征提取算法的2~3倍，方法在采用DAISY算子的同时引入超像素分割，大大提升了图像匹配的效率。结论 本文提出了一种非刚体密集匹配算法，针对非刚体变化的不确定性采用密集特征点进行最优化搜索匹配。实验结果表明，本文的算法对包含小范围非刚性变化的图像匹配上具有较好的适应性，且匹配精度高，视觉效果好，鲁棒性强。
Abstract: Objective Feature-based image matching has always been a fundamental research in the fields of computer vision and pattern recognition. Presently, progress has been made in matching rigid objects. However, in order to facilitate subsequent processing and analyzing, fast and accurate registration of the non-rigid deformation is often necessary in many practical matching problems. At present, the non-rigid matching algorithms at home and abroad are difficult to make perfectly trade-offs in matching precision, speed and robustness. Therefore, it is of great importance to make research on non-rigid matching algorithms which can achieve non-rigid deformation quickly and accurately, and obtain nonlinear transformation parameters with an optimization algorithm. Method Our method is based on the SIFT flow algorithm proposed by Ce Liu et al. for stereo matching. The SIFT Flow algorithm uses fixed-scale SIFT descriptors densely for the entire image lattice, so it cannot well match the scenes containing non-rigid and spatially varying deformations (e.g. the scale and rotation changes). Meanwhile, due to the complex construction process of SIFT feature operators, the algorithm has a unsatisfactory real-time performance. In this study, we introduce the DAISY descriptor to replace the SIFT operator, which not only improves the operator construction speed, but also has a good rotation invariance for its unique circular symmetric structure. This paper presents a non-rigid dense matching algorithm based on DAISY feature descriptor and constrained Patch-Match. Firstly, the DAISY feature descriptor is generated for the reference image and the under-matched image. Secondly, the reference image and the under-matched image are segmented to form super-pixel block structures which are adjacent but non-overlapping. These super-pixel block structures will be used as units for calculating the cost of DAISY feature descriptor in each pixel at the initial position. Then, each pixel in the whole image undergoes propagation followed by a random search based on the Patch-Match algorithm. In the process of random search, the initialization window of the local label is localized and the position tag of each pixel is updated in the space range based on the prior knowledge obtained by pre-process and analysis. The new aggregation cost is calculated through above processes and the position tag with the smaller cost is retained. This process will be repeated until the aggregate cost does not change or the maximum number of iterations is reached. Result Due to the spatially varying deformations of the moving object in traditional optical flow-based stereo matching method, it is necessary to conduct a random search for each pixel in the whole image to provide more matching possibilities. In this paper, three types of images are selected: the standard test sets provided by Middlebury visual website, the coupler buffer images taken with TFDS line array cameras, and the non-rigid images collected by frame cameras. All of these datasets contain non-rigid as well as a small range of deformation. In this experiment, instead of randomly searching the whole image, we perform spatial constraints artificially on the initialization window in the random search of the Patch-Match algorithm, which avoids mismatches caused by the noise or a lack of texture. All of the tests have achieved a better match results. It has been verified by the experiments that the matching accuracy of our method is 86%, and the average matching error of mismatched points is about 5 pixels, which is half of that produced by traditional SIFT flow matching method. And the speed of the DAISY operator adopted in this paper is 1 to 2 times faster than the Dense SIFT in features extraction, which greatly improves the image matching efficiency. Conclusion Traditionally, in order to take into account the matching accuracy of the points with large scale changes, the best matching points need to be searched in the whole image, which can result in mismatches. Aiming at the non-rigid images with small-scale deformation, this paper proposes a non-rigid dense matching algorithm. To deal with the uncertainty of changes in non-rigid images, we use the optimization search algorithm based on dense features. Experiment results indicate that our method is adaptive to the image matching with non-rigid deformation, and successfully achieves a higher matching accuracy and better visual effects than other methods.