Research On Restoration Algorithm For Noised Image Under Complicated Illumination Condition

With the development of science and technology,image acquisition devices can be seen everywhere.These devices are affected by the environment and other factors,the collected images are often there is noise,low contrast,brightness dark,uneven lighting and so on.Usually the traditional recovery algorithm is based on the tensile transformation,but it is easy to produce over-enhancement and distortion.Retinex can recover this kind of images very well,but there are still some weaknesses.For example,it is easy to produce halo,magnify noise and lose details.In this paper,Retinex has been deeply studied and systematically analyzed.With the characteristics of complex noise images taken into consideration,DDDTCWT(Double-Density Dual-Tree Complex Wavelet Transform)de-noising is combined with Retinex transform.In the process of recovering complicated illumination image with noise using fusion algorithm,the main innovations are as follows:1.Aimed at the flaw that threshold estimation is global threshold in the traditional Bayesian threshold de-noising method,an adaptive multi-level and multi-threshold Bayesian threshold estimation method is proposed according to the relationship between the wavelet decomposition level and the noise intensity.2.In order to solve the problem of the pseudo-Gibbs effect or the constant error of the traditional threshold function,a continuous threshold function whose transition is natural and asymptote is y = x has been proposed.3.As for the flaws of complex Gaussian function calculation,fuzzy edge and losing details,a new method is proposed to estimate the illumination image by using the guided filter,which has realized the illumination estimation within the linear time.4.To solve the enhancement and whitening phenomenon in multi-scale Retinex transform process,a Multi-Scale Retinex transform method which combines image grey entropy,with adaptive weighting coefficient has been proposed.In this paper,we experiment and verify the low-contrast noise-free complex illumination images.A series of methods,such as PSNR(Peak Signal-to-Noise Ratio)?MSE(Mean Squared Error)evaluation?(SSIM)Structural Similarity evaluation? information entropy evaluation? clarity evaluation and so on,are compared with the classical method.The experimental results show that the proposed method can effectively enhance the detail,edge and texture of complex illumination images,reducing the noise and improving the contrast,solving the defects of traditional Retinex recovery image noise amplification and flare as well.The algorithm described in this paper can be widely used in the recovery of monitoring images,aerial images and so on,which has a very large social value.