Lighting Variability Correction for Pill Identification
Seungtae Kang, Gil‐Jin Jang, Minho Lee · 2018
This paper proposes a novel method for the compensation of the illumination variations. To find accurate approximate of the shading conditions, class activation map (CAM) obtained by the output of convolutional neural networks (CNNs) provides weights for the shading parameter estimation. We applied the shading compensation to the pill images, and obtained improved surface images.