Abstract <p>Image matching is at the base of many image processing and computer vision problems, such as object recognition or structure from motion. Current methods rely on good feature descriptors and mismatch removal strategies for detection and matching. However, the common used features are hand-crafted, local low-level information and do not include middle or high level information, thus cannot apply to different scenarios. Image features extracted by a convolutional neural network have achieved the state-of-the-art performance for image classification and retrieval problems. In this paper, we proposed a robust image match approach based on deep convolutional neural network features and Delaunay triangulation for mismatch removal to improve match accuracy. Firstly, given a pairs of image, detecting MSER points and extracting corresponding patches; secondly, based on Siamese network model, extracting Siamese Network descriptors and obtaining initial matches by L2 Euclidean distance matching; then refining the initial matches by angle constraint; finally, refining matches by Delaunay triangulation to get the final matches. The experiment results demonstrate that our method is efficient and robust.</p>