| Recently,face recognition has been widely used in various fields such as public security management,e-commerce,education,and healthcare due to its non-contact,convenient and fast.However,face recognition in low-light environments is still a challenging task,including:(1)the low quality of low-light face images,which require improving their visibility and visual quality;(2)low-light face images are susceptible to damage from factors such as occlusion,pose and expression changes,it is necessary to improve the robustness to damaged low-light face images.To address these problems,this paper is dedicated to improving improve the performance of face recognition systems in low-light environments from two perspectives:low-light image enhancement preprocessing and face recognition algorithms,which is helpful to related disciplines and commercial applications.The main contents of this paper are summerized as follows:1.To solve the issue that existing Retinex-based algorithms often neglected the dense noise or removed by pre/post-processing steps,this paper proposes a robust lowlight image enhancement algorithm(LLERM)based on structure-aware and low-rank embedded Retinex model,which integrates low-light image enhancement and denoising into a unified framework,achieving noise suppression throughout the entire process.Speaifically,Starting with a robust image reconstruction metric via l2,1 norm,LLERM embeds low-rank prior based on image patch and a new adaptive illumination adjustment method into the Retinex decomposition process,where the former achieves noise suppression while revealing texture structure,the latter maintains the balance between brightness and naturalness.Moreover,an accelerated low-rank approximation strategy is designed to reduce the complexity.And a sequential optimization method is developed to solve the model and address the mutual interference of noise between the components.Extensi ve experimental results demonstrate the superiority of LLERM in low-light face images enhancement.2.To solve the issue that low-light face images are inevitably damaged by noise such as illumination,occlusion,posture and expression changes,and the existing face recognition methods neglected the the correlation between features,this paper proposes a Joint Latent Low-Rank and Non-Negative Induced Sparse Representation(JLSRC)for face recognition.Specifically,JLSRC adaptively learns two clean low-rank reconstructed dictionaries jointly via an extended latent low-rank representation and then embeds a non-negative constraint and an Elastic Net regularization in the coefficient vectors of the dictionaries to enhance the performance on classification.In this way,the learned low-rank dictionaries can be mutually boosted to extract discriminative features and handle the noise,and the obtained coefficient vectors are simultaneously both sparse and discriminative.Moreover,the proposed method seamlessly and elegantly integrates low-rank learning and sparse representation-based classification.Extensive experimentson several challenging face datasets have demonstrated the effectiveness and robustness of JLSRC in low-light face recognition tasks.In summary,the proposed LLERM achieves high-quality enhancement of lowlight face images,while JLSRC addresses the degradation of low-light face recognition performance caused by occlusion and other noise factors.These two algorithms complement each other,and have achieved excellent experimental results on multiple lowlight face image datasets.These findings provide a systematic solution for addressing low-light face recognition challenges in practical applications... |
