Study On Space Motion Image Denoising And Enhancement

The quality of space motion image is generally not high due to many constraints such as space radiation and shooting environment. Spatial image denoising and enhancement technologies have significant theoretical significance and application value. This thesis devotes to study the moving image denoising and enhancement for solving the problems such as noise pollution and low illumination. The main work is completed as follows:(1) A noise classification algorithm based on support vector machine (SVM) was proposed (SVM-NC). The proposed algorithm can find out the hyper-plane which can classify the motion image affected by the specific noise through high-dimensional mapping of eigen-values. The image feature vectors are extracted to analyze the relation between the image quality and the eigen-values when the individual image is affected by different categories and varying degrees noise. A blind image quality assessment method based on noise classification was proposed (NRNC-IQA). By evaluating the quality of de-noising and enhancement by NRNC-IQA and modifying the parameters, NRNC-IQA predicts the spatial image quality effectively. The experimental results showed that, compared with PSNR-NC and BIQI-NC algorithm, the proposed SVM-NC algorithm improves significantly in accuracy index; NRNC-IQA algorithm can basically satisfy the no reference image objective evaluation standard requirements.(2) The image de-noising algorithm based on SSIM total variation (SSIM-TV) was proposed. The algorithm classifies the noise sources, and effectively quantifies the influence degree of each noise source. By means of calculating different noise sources and the specific weights based on different noise sources, all noise factors can be adjusted, and consequently adaptively de-noising can be employed for different space noise image sequences. The experimental results showed that compared with MSE-TV, MFD and WFD algorithm, the different noise variance, PSNR, RMSE and RSSIM of the proposed SSIM-TV algorithm are all improved significantly.(3) The low illumination Retinex algorithm based on guided filtering was proposed. The proposed algorithm makes use of bilateral filtering to estimate the illumination of input image, and then optimizes the illumination by the guided filtering, which can recover the shaded area. In the meantime, also maintains the highlighted details of the image. The proposed algorithm effectively eliminates the halo artifacts and over-enhancement phenomenon. Compared with Retinex-G and Retinex-B algorithm, the average, standard deviation and entropy for the proposed Retinex-LG algorithm are all improved significantly.(4) The motion image denosing and enhancement system was designed and developed. The proposed SVM-NC, SSIM-TV and Retinex-LG algorithms were all proved by the system and system test was also carried on.