High-Dynamic Range, Night Vision, Image-Fusion Algorithm Based on a Decomposition Convolution Neural Network

The dynamic range of night vision scenes is typically very large. Owing to the limited dynamic range of the traditional low-light-level imaging technology, the captured images are always partially overexposed or underexposed. Multi-exposure fusion is the most effective method for overcoming the dynamic range limitations of sensors. Recently, deep learning has achieved tremendous progress in many fields. However, only a few breakthroughs have been reported on high-dynamic image fusion with the deep learning method. Additionally, many problems have been reported in conjunctions with commonly used deep-learning methods. In this study, a high-dynamic image fusion algorithm is proposed based on the decomposition convolution neural network and weighted sparse representation. Based on image decomposition, the problem of the acquisition in training samples in network training can be solved. Therefore, the classification accuracy of the network can be improved. Additionally, the decomposition structure reduces the workload of each layer and improves the efficiency and quality of the image fusion outcome.