Droplet Image Reconstruction Based on Generative Adversarial Network

Abstract

Abstract In the digital microfluidic experiments, the improper adjustments of the camera focus and background illumination lead to the phenomena of low illumination and blurred edges in the droplet image, which seriously interferes with information acquisition. Removing these blurred factors is an essential pretreatment step before information extraction. In this paper, a generative adversarial network model combining multi-scale convolution and attention mechanism is proposed to reconstruct the droplet image. The feature reconstruction module in generator can reconstruct the image feature maps from multiple scales. The fusion module is used to fuse the multi-scale feature maps into a reconstructed sharp image. The new model was trained on the data set which was made by the Style Transfer. Experimental results show that the proposed model can significantly improve the visual quality of images, effectively reduce the blur and improve the background illumination.