Abstract:

This paper discusses a technique for sharpening images in the dark to support museum learning. Museum learning, which is an opportunity for visitors to experience culture and develop various interests, is becoming more and more important year by year. In response to this trend, various research and development efforts have been made for supporting technologies for museum learning. One of these assistive technologies is the automation of guided tours for museum visitors. For this automation, robots have been developed that use cameras to detect exhibits and guide visitors accordingly. However, the lighting in museums is suppressed to protect exhibits, and in such dark environments, the color and shape of objects become blurred, which reduces the accuracy of detection.
To address this issue, the authors proposed a method to sharpen dark images using deep learning based on human visual characteristics. Human vision perceives object colors independently of illumination by separating light incident on the eye into illumination and reflected light components. In the proposed method, the reflectance component, which represents object-specific color, is considered as a sharpening image, and the illumination component is estimated using deep learning, and this component is separated from the dark image to achieve sharpening. In the evaluation experiment, the sharpened image was evaluated in terms of both image quality and detection accuracy. First, we confirmed that the proposed method can sharpen images with high image quality by calculating the similarity between the sharpened image and the light image. Next, we evaluated the detection accuracy of the proposed method by performing object detection on the dark image and the sharpened image. The results showed that the proposed method is effective in improving detection accuracy by sharpening.

Keywords:

Museum learning support, Low-light image enhancement, Object detection.