基于量子k-means聚类的图像复制粘贴检测

Image Copy-Paste Detection Based on Quantum K-Means Clustering

在同幅图像复制粘贴检测过程中,提取出的图像特征关键点数量庞大且特征维度多,导致计算复杂度高、特征匹配时间消耗大、鲁棒性差等问题。为克服上述问题,运用量子计算的并行性,引入量子k-means聚类算法对特征关键点进行特征匹配实现图像复制粘贴检测。上述方法先根据尺度不变特征变换提取图像关键点特征子描述向量,然后将特征点描述向量做酉变换和黑箱Oracle操作进行量子化处理,再对其进行量子相似性查找进行聚类,最后对聚类特征关键点进行特征匹配得到图像复制粘贴区域。实验结果表明,量子k-means特征匹配算法在图像检测中降低了时间消耗且具有较好的鲁棒性,即使被复制区域经过旋转、缩放、模糊、亮度修改、JPEG压缩和噪声添加等图像处理,上述方法仍能准确地匹配到复制粘贴区域特征点。

In the copy-paste detection process of the same image,the extracted image feature key points have a large number of feature dimensions,resulting in high computational complexity,feature matching time consumption,poor robustness and other problems.In order to overcome the above problems,quantum computing parallelism was used to introduce the quantum K-means clustering algorithm for feature matching of key points to achieve image copy-paste detection.The above method first extracted the feature sub description vector of image key points based on scale invariant feature transformation,and quantized the feature point description vector through unitary transformation and black box Oracle operation,then performed quantum similarity search for clustering,and finally performed feature matching on the clustered feature key points to obtain the image copy and paste area.The experimental results show that the quantum K-means feature matching algorithm reduces the time consumption and has good robustness in image detection.Even if the copied region is processed by rotation,scaling,blur,brightness modification,JPEG compression and noise addition,the method can still accurately match the feature points of the copy-paste region.