Robustness of Reflection Symmetry Detection Methods on Visual Stresses in Human Perception Perspective

Symmetry is one of the most frequently observed fundamental regularities in the visual characteristic of real-world objects. The human brain has been trained to respond quickly to symmetry patterns, organizing them as salient clues for the unique description of objects. Recently, automatic symmetry detection methods have been widely introduced in computer vision and graphics fields for 2-D and 3-D object data, including reflection, translation, and rotation symmetry patterns. Researchers have invented features inspired by a human vision system and have adopted deep learning approaches. On the other side, traditional performance evaluations have been conducted on a unified test data set containing random degrees of diverse visual challenges. However, they ignore observing the insight of usability and practicality of the methods in higher level tasks, such as object recognition. In this paper, we carefully organize the visual stress data set for reflection symmetry detection evaluation proposing a novel evaluation framework. The state-of-the-art reflection symmetry detection methods are re-evaluated and analyzed in human perception perspective.