Computational lighting for extracting optical features from RGB images. (February 2020)

• Record Type: Journal Article
• Authors: Higashi, Hiroshi; Bui, Minh Vu; Aziz, Ahmad Syahir Bin; Nakauchi, Shigeki
• Is Part Of: Measurement. Volume 151(2020)
• Publication Date: 2020-02
• Subjects: Optical measurements -- Optical features -- Computational lighting -- Spectrum estimation -- Machine learning; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2019.107183
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: Illuminant design by machine learning can handily predict optics-related properties. Low-cost implementation achieved by using commercial LEDs and RGB cameras. Our method does not have to use any spectral information. Synthesized illuminants reduce the measurement time. Abstract: Optical measurements for capturing optical features that show the physical and chemical properties of target objects and scenes fall under the nondestructive measurement method. These measurements require a long period of time and expensive specialized equipment. This paper proposes a practical system composed of commercial LEDs and RGB cameras for extracting optical features and predicting the properties from RGB images. Besides the predictor optimization by supervised learning, the system also utilizes computational lighting techniques for optimizing the synthesized illuminants, which are composed of readily available LEDs. In addition to the low-cost implementation, our system provides fast measurement because the number of images that are photographed can be reduced through computational lighting. We demonstrate the effectiveness of our system in prediction problems where we analyze the fluorescence intensity of scenes drawn with markers and pearl quality.