Research And Application Of Improved Retinex Enhancement Algorithms

With the improvement of social informatization and the optimization of digital imaging equipment,image enhancement technology has increasingly penetrated into optical detection,medicine,aerospace and other fields.No matter Optical interference fringe patterns in optical detection field,Optical Coherence Tomography(OCT)images in medical field or foggy images in aerospace field,the image contrast is low.And the optical interference fringe patterns and OCT images are affected by speckle noise.Therefore,it is necessary to enhance these images.Image enhancement algorithm based on Retinex theory is one of the effective methods of image enhancement technology,which has the advantages of contrast improvement,color constancy and color fidelity.In this thesis,the Retinex theory is deeply studied for the problems of optical interference fringe patterns,retinal OCT images and foggy images,and improved algorithms are proposed based on Retinex theory.The contributions of this thesis can be summarized into three aspects as follows:(1)To address the problem of low contrast and uneven illumination in optical interference fringe patterns,the multi-scale Retinex enhancement algorithm is introduced into the field of optical filed.By using the multi-scale Retinex algorithm to reduce the influence of incident light and obtain the essential information of the image,the problem of uneven illumination in the image can be solved while improving the contrast.Experimental results show that compared with the commonly used histogram equalization algorithm and the differential equation algorithm,the proposed algorithm can effectively improve the image contrast and eliminate the uneven illumination.(2)To address the problem of low contrast and speckle noise in retinal OCT images,a selective Retinex enhancement algorithm based on Fuzzy c-means(FCM)clustering is proposed.This algorithm obtains new feature images through feature extraction and fusion,and uses FCM clustering algorithm to cluster and obtain the target mask.The mask is combined with Retinex algorithm to selectively enhance the image Finally,Block-matching 3D(BM3D)filtering method is used to filter the enhanced images.Experiments show that compared with the results of commonly used BM3 D algorithms and Shearlet transform,the proposed algorithm can simultaneously enhance the structural information and suppress the background noise,and maintain the image texture information better.(3)To address the problem of image detail information retention during foggy image enhancement,a Retinex image defogging algorithm based on variational image decomposition is proposed in this thesis.The algorithm decomposes the image into a cartoon part and a texture part,and performs dehazing treatment on the cartoon part.Experimental results show that compared with the classical histogram equalization method,multi-scale Retinex algorithms and Dark Channel defogging algorithm,our method can keep the detailed information better while defogging.Then,the proposed method is improved and applied to underwater image enhancement.Underwater image is converted from RGB space to HSV space for separate processing,which can effectively improve the color deviation problem of underwater images.