PROJECT TITLE :

Deep Guided Learning for Fast Multi-Exposure Image Fusion

ABSTRACT:

MEF-Net is a rapid multi-exposure image fusion (MEF) approach for static image sequences of adjustable spatial resolution and exposure number that can be used to a variety of datasets. Initially, we use a low-resolution version of the input sequence to forecast the weight map. The weight maps are then resampled using a guided filter. Based on weighted combination, the final image is generated. MEF-Net is trained end-to-end by improving the perceptually calibrated MEF structural similarity (MEF-SSIM) index over a database of full-resolution training sequences. Optimized MEF-Net consistently improves visual quality and runs 10 to 1000 times quicker than state-of-the-art approaches on an independent set of test sequences.