Attention-based Low-light Image Enhancement Algorithm

In a low-light environment,due to low photon energy and insufficient sensitivity of imaging equipment,the captured images often have problems such as low signal-to-noise ratio,insufficient contrast,and severe loss of detail information.Such images will not only bring poor appearance but also affect the execution of some high-level computer vision tasks,such as target detection,semantic segmentation,and pedestrian re-recognition.Therefore,the low-light image enhancement algorithm has excellent practical application value.Designing a more reliable low illumination image enhancement algorithm is the leading research content of this paper.Firstly,this paper introduces the traditional image processing technology,the background knowledge of deep learning,and the main research content of this paper.Then,using the traditional image processing theory,the low-light enhancement task is analyzed from the perspective of the frequency domain.Finally,a low-light image spectrum enhancement algorithm based on the attention mechanism is proposed according to the low-light image spectrum characteristics.The main work and innovations of this paper are as follows:a lowlight image enhancement algorithm based on frequency domain attention mechanism is proposed,which uses a learnable neural network to enhance and denoise the low-light image in the frequency domain instead of the manual design of traditional filter;A low-light image enhancement algorithm based on hybrid attention mechanism is proposed,which reduces the information loss caused by global average pooling,makes the network have more vital feature expression ability,can more fully learn the importance of different feature maps,and enhance the performance and generalization ability of the algorithm.In this paper,many experiments are carried out on SID datasets.The results show that the algorithm proposed in this paper has a better enhancement effect than the existing low-light image enhancement algorithms.At the same time,ablation experiments confirmed the effectiveness of the attention module proposed in this paper.Finally,this paper tests on the real-world dataset,and the test results show that the algorithm designed in this paper has strong practicability and generalization.