PROJECT TITLE :

Learning a Deep Single Image Contrast Enhancer from Multi-Exposure Images - 2018

ABSTRACT:

Due to the poor lighting condition and limited dynamic vary of digital imaging devices, the recorded images are typically underneath-/over-exposed and with low contrast. Most of previous single image distinction enhancement (SICE) strategies regulate the tone curve to correct the contrast of an input image. Those strategies, however, often fail in revealing image details as a result of of the limited info in a single image. On the opposite hand, the SICE task will be better accomplished if we tend to can learn extra data from appropriately collected training information. During this Project, we propose to use the convolutional neural network (CNN) to coach a SICE enhancer. One key issue is how to construct a training data set of low-distinction and high-distinction image pairs for finish-to-finish CNN learning. To this finish, we have a tendency to build a giant-scale multi-exposure image knowledge set, which contains 589 elaborately selected high-resolution multi-exposure sequences with four,413 images. Thirteen representative multi-exposure image fusion and stack-based high dynamic vary imaging algorithms are utilized to get the distinction enhanced images for every sequence, and subjective experiments are conducted to screen the best quality one because the reference image of each scene. With the made information set, a CNN will be simply trained because the SICE enhancer to enhance the distinction of an below-/over-exposure image. Experimental results demonstrate the advantages of our method over existing SICE ways with a important margin.