A Benchmark for the Evaluation of Corner Detectors

Abstract

Corners are an important kind of image feature and play a crucial role in solving various tasks. Over the past few decades, a great number of corner detectors have been proposed. However, there is no benchmark dataset with labeled ground-truth corners and unified metrics to evaluate their corner detection performance. In this paper, we build three benchmark datasets for corner detection. The first two consist of those binary and gray-value images that have been commonly used in previous corner detection studies. The third one contains a set of urban images, called the Urban-Corner dataset. For each test image in these three datasets, the ground-truth corners are manually labeled as objectively as possible with the assistance of a line segment detector. Then, a set of benchmark evaluation metrics is suggested, including five conventional ones: the precision, the recall, the arithmetic mean of precision and recall (APR), the F score, the localization error (Le), and a new one proposed in this work called the repeatability referenced to ground truth (RGT). Finally, a comprehensive evaluation of current state-of-the-art corner detectors is conducted.

Revisiting Mehrotra and Nichani’s Corner Detection Method for Improvement with Truncated Anisotropic Gaussian Filtering

“…Finally, in the proposed experiments, the number of detected corners (card(D c )) is the same as the number of the true corners in the ground truth (card(T c )); consequently, the compared corner detection methods extract the same number of corners in each image. Note that the RMSE is also also called "localization error", see [24]. It does not penalize the corners detected very close to their reference, unlike precision/recall-type metrics, which do not tolerate small pixel deviations.…”

“…The appropriate choice of such measures directly impacts the algorithm's performance, influencing its sensitivity to noise and ability to handle scale and orientation variations. There are different approaches to determining the cornerness measure by direct computation using filtering techniques; a recent review [22] details these measures, and can be further complemented with [23,24].…”

“…It provides a historical perspective and helps readers understand the significance of our contributions. While newer corner detection methods [34][35][36] based on deep learning and other advanced techniques have emerged (the introduction by Zhang et al [24] is a nice compendium), the classical algorithms by image filtering like Harris [32] and Shi-Tomasi [33] remain relevant because they provide a solid baseline for performance and continue to be effective in many practical applications. Additionally, they are often the first choice when simplicity, efficiency, and well-understood behavior are critical.…”

Vectorizing binary image boundaries with symmetric shape detection, bisection and optimal parameterization

“…Corner point detection has been extensively investigated in existing research studies, and various corner point detection methods have been developed. Recently, Zhang et al made a comprehensive evaluation of state‐of‐the‐art corner point detection methods and identified several corner point detection methods with top performance 34 . In our following work, we will evaluate these identified methods and implement a more suitable corner point detection method to avoid parameter adjustment.…”