Research On Image Denoising Of Low Light Level

Image denoising is one of the most important steps in image proeessing. It' s purpose is to enhance the SNR between original image and de-noised image, improve the characteristics of image. The characters low light level image is weak contrast gradient and low SNR. All low light level image processing revolves to strengthen the contrast gradient and to enhance the SNR. Image denoising methods can be classified as the time domain and the frequency domain methods. In essence, however, they both make use of different characteristics of the noise and the signal in frequency domain. There is a dilemma between image denoising and image detail reservating. General filtering methods based the Fourier transform can' t enhance the spatial resolution and improve the SNR simultaneity. Although low-pass filtering can eliminate the noise by smoothing technology, it makes the image edges fuzzy. The image edges will be steeper by high-pass filtering. However, he background noise is also enhanced at the same time. The signal and noise can be differentiated more when the wavelet Transform that is a powerful tool for signal analysis emerges. For examplethe signal and noise can be differentiated by using the eddy information since the image edge plays a dominating role in the vision.In this dissertation, those denoising algorithms are studied for low light level image, such as that by Mathematics Morphology, that in frequency domain or spatial domain, that on wavelet transformation and so on. The first method includes corrosion and expansion; The second in frequency domain comprises low-pass filtering and homomorphic filtering; It contains the median filtering and the average filtering in spatial domain;the method of wavelet transformation covers linear filtering, denoising by threshold or by wavelet model;and that of partial differential equation uses linear diffusion, nonlinear scalar or tensor diffusion. Above all these algorithms, we focus on analyzing wavelet transformation. To begin with, we simulate special algorithm in matlab to deduce the artificial noises on system images. Then we turn to our new images and do the similar simulation with proper tools. Finally, we get some result and compare data, according to different types of wavelet in this process.