Light consistency correction for liquid crystal tunable filter hyper-spectral imaging system

• jianxin zhang, Yupeng Zhang, Miao Qian, and Xinen Zhang
• received 12/12/2023; accepted 04/18/2024; posted 04/19/2024; Doc. ID 515706
• Abstract: In hyperspectral images, every pixel encompasses continuous spectral information.Compared with traditional colorimeters, using hyperspectral imaging systems (HIS) for fabriccolor measurement can obtain richer color information. However, measuring fabric colorswith Liquid Crystal Tunable Filter (LCTF) HIS encounters challenges related to light consistency. In this paper, we adopted an innovative approach, integrating Gradient BoostedDecision Trees (GBDT) with a sliding window algorithm to develop a uniformity calibrationmodel addressing the illumination uniformity issue. To address the consistency issues acrossvarious light sources, we further adopted a deep neural network (DNN) model to correct thereflectance measurements under different light sources. Subsequently, this model was mergedwith the uniformity calibration model to form a light consistency correction model. Throughcalibration, we successfully reduced the color difference of the corrected samples from 3.636to 0.854, an enhancement of 76.51%. This means that after calibration we can achieveconsistency in fabric color measurements under non-uniform lighting and different lightsources.