Research On Dehazing Algorithm Based On Prior Constraints And Neural Network

The rapid advancement of AI has had a profound effect on our daily lives,infiltrating multiple industries and fostering social and economic progress.Of these,computer vision is an indispensable factor,able to imitate biological vision to comprehend and manage visual tasks,such as intelligent transportation,payment security,medical image analysis,etc.Among them,computer vision is an indispensable component,which can mimic biological vision to better understand and process visual tasks,such as intelligent transportation,payment security,medical image analysis,etc.However,like the human eye,in outdoor adverse weather conditions,it is easy to lead to reduced visual performance,reduced visibility or even failure.In order to reduce the impact of bad weather(fog,haze,etc.)on a series of advanced computer vision tasks such as object detection,image recognition,etc.,it is necessary to obtain highquality clear images through image dehazing as preprocessing.Therefore,it is of great significance to study concise and efficient image dehazing algorithms to ensure the smooth progress of advanced visual tasks.In this paper,the existing image restoration algorithm based on physical model and the image restoration algorithm based on data drive are analyzed,the advantages and shortcomings of the algorithm are summarized,and an improved algorithm is proposed from the above two perspectives for its shortcomings.Single-image defogging is widely used in outdoor optical image acquisition equipment.For the problems of incomplete defogging and color bias of existing methods,this paper proposes an atmospheric light curtain estimation method.Firstly,the low-frequency components of the foggy image are separated by discrete cosine transformation,and then the linear relationship between the low-frequency components and the atmospheric light curtain is used to propose a Gaussian function to establish an adaptive atmospheric light curtain acquisition model to obtain the initial atmospheric light curtain.Then,by constructing a linear function with unknown parameters,combined with the local atmospheric light exploration method,and applying the average saturation prior to the atmospheric scattering model,a global optimization function is obtained to obtain the unknown parameters,and finally,a clear fogfree image is obtained by the atmospheric scattering model.Experiments show that compared with some classical and advanced algorithms,the algorithm has strong versatility,clear restoration results,natural color,and can be better than other comparison algorithms in some objective evaluation indicators.The existing physical model-based dehazing algorithms are easily limited by the model and the limitations of convolutional neural networks(CNNs)in the expression of global information of images,and general CNN-based dehazing models are often difficult to balance the local information and global information of images,and the robustness is poor.Therefore,a gated fusion dehazing network combining convolutional neural network and Transformer is proposed to achieve end-to-end dehazing.The network model consists of two main modules:the Feature Extraction Module(FEM)and the Color Recovery Module(CRM).Among them the feature extraction module is composed of a local feature extraction module(LFEM)composed of CNN,a global feature extraction module(GFEM)composed of Transformer,and a gated fusion module(GFM).The feature extraction module is responsible for extracting the local and global features of the image,and then fusing the features through the gated fusion module,and then using this feature extraction module as the basic block to form the entire network model.Finally,the CRM module restores the clear image and restores the color.Experiments show that the algorithm has good recovery results and is competitive in both subjective and objective evaluation,which verifies the effectiveness of the algorithm.