基于动态感知曲线调整策略的自适应低光照图像增强方法OA

In order to solve the shortcomings of the existing low-light enhancement methods in complex lighting scenes,such as limited generalization ability and insufficient adaptability of parameter adjust-ment,a reference-free low-light image enhancement method was proposed based on dynamic perception curve adjustment strategy(DPCAS).The lightweight feature perception module(LFPM)and dynamic curve adjustment module(DCAM)were included in DPCAS.The lightweight architecture and multi-scale dual-attention mechanism were used in LFPM to realize feature extraction and aggregation,and the parameters of the mapping curve were adjusted adaptively in DCAM with dynamic iteration mechanism.The dynamic balance between local detail enhancement and global exposure correction was achieved with the two modules.Besides,a two-stage training mechanism was proposed by adaptively adjusting weights of multiple loss functions,so as to achieve more stable convergence and more accurate target optimization control.Qualitative and quantitative experiments on the testing sets with reference and reference-free sets proved that our method was superior to most existing low-light image enhancement methods in terms of subjective quality,and had higher level for the quantitative indicators such as PSNR,SSIM and NIQE,and had high operational efficiency.The LFPM,DCAM and adaptive adjustment mechanism for loss functions were also verified to be effective by ablation experiments.