Abstract
In the current article, an improvement to an existing state-of-the-art RGB-NIR based multispectral imaging technique is proposed. In the previous research work it was learnt that use of a greater number of clusters can provide better results, but at the cost of increased execution time. The current work therefore builds on the previous works by utilizing new structures which will help in decreasing fusion time. Also proposed in the current article is an extension of this fusion technique to parametric scene parsing. For this purpose, about 4000 RGB-NIR image pairs are obtained from an existing dataset and performed the above fusion procedure to create a new corpus consisting only of enhanced fused images. These images are used as training set for the current deep learning based semantic segmentation architecture. Benefits on two fronts are observed: A) Image fusion now has better contrast, sharpness, enhanced visibility, reduced execution time. B) Semantic segmentation results in superior performance as compared to other techniques based solely on RGB images. Keywords: Image fusion, multispectral imaging, RGB, Near infrared.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have