Machine Learning on Topological Constraint for Mismatching Removal.

Chentao Shen,Zaixing He,Xinyue Zhao,Mengyu Qiu
DOI: https://doi.org/10.1145/3607822.3616410
2023-01-01
Abstract:Image feature matching is a critical task in human-computer interaction, such as driverless cars, collaborative medical robots, aiming to establish correspondence between two images inputted from human or computer. To ensure the accuracy of matches, removing mismatches is of utmost importance. In recent years, machine learning has emerged as a new perspective for achieving effective mismatch removal. However, current learning-based methods suffer from a lack of generalizability due to their heavy reliance on extensive image data for training. Consequently, handling cases with a high mismatch ratio becomes challenging. In this paper, a novel approach is proposed that incorporates the triangular topology constraint into machine learning, which we called LTM. By summarizing topology constraints around the matching points and integrating the idea of sampling, we successfully address the mismatch removal task. Notably, our method stands out by requiring few parameters as input, enabling us to train it using just several image pairs from four sets. Remarkably, our method achieves outstanding results on diverse datasets using various machine learning techniques compared with many existing methods.
What problem does this paper attempt to address?