基于卷积特征提取和DTW的图像相似度评估算法

Image Similarity Evaluation Algorithm Based on Convolution Feature Extraction and DTW

计算机视觉领域中时常需要评估两张图片的相似度。已有的方法或受到人工特征提取的限制,或需要特殊的数据集。考虑到它们的局限性,通过将目标检测网络的卷积层与二维动态时间规整结合起来,给出一种无需特殊数据集和人工特征提取的图片相似度评估算法。

In the field of computer vision,it is often necessary to evaluate the similarity of two pictures.The existing methods may require manual fea⁃ture extraction or special data sets.Considering their limitations,by combining the convolutional layer of the target detection network with two-dimensional dynamic time warping,a picture similarity evaluation algorithm without special data sets and manual feature extraction is given.