| The low light level images are different from general visible light images. It is formed by the process of photoelectric conversion and electron multiplier. Due to the low illuminance and poor background, there are obvious random flicker noises superimposed on the output images. The lower the illuminance is, the more serious the noises are. At the same time the brightness and contrast also decreased. Thus the obtained information is of very low signal to noise ratio. So the output images are without enough resolution and contrast to be observed and identified. Obviously, the noise is the key factor that influences the performance of the low light level imaging system. Therefore it is of great practical significance to investigate the noise characteristics to improve the image quality, enhance the signal to noise ratio, promote and expand the application fields.In this paper, based on the image noise of the low light level ICCD, the characteristic of the noise was studied. The experimental research on denoising using the existing ditital image processing method were also be done in order to improve the quality of the image. During the research, much attention was paid not only to the theory and innovation, but also to the actual operation availability. All of these work has great important theoretical and practical significance on the development of noise theory and image quality of the low light level imaging system.The main research is as follows:(1)The structure, working mechanism, together with the sources and characteristics of the noise were discussed. And model of the noise was established from a unified perspective between the time domain stochastic process and the space domain of stochastic process. It was indicated that the noise sources come from imaging system and CCD itself. The noise can be divided into two kinds as far as the final forms of the noise concerned. One kins is the electronic noise, such as the incident photon noise, the quantum fluctuation noise in the photoelectric conversion process, the dark noise from the photoelectric cathode, the noise from MCP detection efficient and gain fluctuation. This kind of noise is randomly fluctuate and determined by the inherent statistical nature of the changes of the photons and electrons with time and space.This noise shows on the screen as small and faint granular twinkles and cannot be eliminated. Only though the image processing method, this kind of noise can be inhibited. Another kind of noise is the ion noise and come from the free ions in the device, such as the absorbed ions in the internal of the device and the electrode, these ions has a dynamic process between the adsorption and desorption. This kind of noise shows on the screen as a large, bright random twinkles. This noise is not inherent an can be eliminated by improving the quality of photoelectric cathode, MCP, electrode formation process and the vacuum.(2) Based on the mean filtering method, the reference denoising method was used to improve the image quality of the low light level. In this method, firstly, the absolute value of the difference between the pixel gray of the filter window and the mean value of all pixels gray, then compared with the threshold value and the final pixel gray value can be ultimately determined.(3) In the experiment, a kind of faster algorithm by using the multiple one-dimensional median filter to replace the2D median filtering was introduced, and a improved extremum median adaptive denosing method was adopted to improve the image quality in this paper. In this method, firstly, the pixels was divided into noise and signal according to the judging standard, then the noise point was replace by the mean value of neighborhood according to the spatial correlation principle. At the same time, the filter window size can be automatically changed according to the change of conditions. So the detail of the image can be well protected.(4) The value filtering technology is the dominant ideology in mathematics and statistics on the occurrence probability of the maximum gray value can be used as the result. By using spatial random of the noise,the frame integral method is to sum up the corresponding pixel gray value from two images with different time. In this paper, this method and the method of frame integral were combined in the experimental study and achieved good results.(5) Based on the difference image of multi dimensional generalized morphological-difference filter. E.g. based on multi dimensional morphological filter, a group difference image containing different sizes of noise and image information was obtained. By denoising using difference and sum up with the final morphological fileter, the targer image not only trtain the image details but also remove much noise.(6) The application of the wavelet analysis on image denoising was studied. By forming a new class of threshold function combining the " hard threshold " with " soft thresholding ", the image noise can be removed. Firstly, decomposing the image using a layer of Haar wavelet, then processing the new combination of soft and hard threshold threshold function on horizontal, diagonal and vertical direction of the high frequency wavelet coefficient threshold. Finally reconstructing the wavelet coefficient, the image denoising results can be shown..The innovation in this paper is as follows:(1)On the basis of the value idea and frame integral method, the frame integral method is put forward. It is indicated that this method can be inhibit the bright noise and dark noise of the low light level image. At the same time, the protection performance on the image edge is better than the frame integral method and of great application value. The PSNR is29.32and35.57for these two methods, respectively.(2) According to the principle of generalized morphological,differential filter and difference image, an improved generalized morphological-difference image denoising method was proposed. It not only eliminates the noise but also keep details of the target image. The generalized morphological-differential filtering for the processed image is higher than average frames of image filter processing. The PSNR of the laster is34.57, while the former is41.52. (3) In order to the shortcoming of the soft and hard thresholding methods, a new thresholding function was constructed by combining the sofe and had thresholding methods and used to denoising on the low light level image. This method has good continuity and convenient adjustment, the final image SNR has been further improved, and has higher application value. The PSNR is37.49,44.76and44.81for the frame integral method, hard threshold method and soft threshold methed, respectively. While the PSNR is45.87for the algorithm adopted in this paper. |
PDF Full Text Request