CNNs for Image Region Comparison via Learned Similarity Functions
Yu Wang, Nozomu Tanaka, Ruijia Chen · 2023
In this paper we show how to learn directly from imagedata (i.e., without resorting to manually-designed features)a general similarity function for comparing image patches,which is a task of fundamental importance for many computer vision problems. To encode such a function, we optfor a CNN-based model that is trained to account for awide variety of changes in image appearance. To that end,we explore and study multiple neural network architectures,which are specifically adapted to this task. We show thatsuch an approach can significantly outperform the state-ofthe-art on several problems and benchmark datasets. Comparing patches across images is probably one of themost fundamental tasks in computer vision and image analysis. It is often used as a subroutine that plays an importantrole in a wide variety of vision tasks. These can range fromlow-level tasks such as structure from motion, wide baselinematching, building panoramas, and image super-resolution,up to higher-level tasks such as object recognition, imageretrieval, and classification of object categories, to mentiona few characteristic examples.