Sunset region detection and interpretable evaluation metrics using the YCbCr color space for camera imagery

Abstract

In recent years, tourism demand for natural landscapes has been increasing. Among natural landscapes, this study focuses on sunset scenery and proposes a method to evaluate their visual features. Existing sunset detections and prediction methods mainly rely on a single index, making it hard to understand actual characteristics of sunset imagery. To address this issue, this study developed a sunset image detection -method using the YCbCr color space, and proposed three interpretable features for evaluating sunset imagery: sunset area ratio, red color intensity, and shape complexity. Applying the proposed approach to one-year fixed-point camera images at Chiba University confirmed that our method can classify sunset images appropriately. Furthermore, sunset images with atypical values in the feature space were found to exhibit visually distinctive characteristics, suggesting that the proposed features would also be useful for identifying rare sunsets. The method developed in this study is expected to contribute to more advanced landscape prediction and generation tasks, such as sunset scenery forecasting.