Image Restoration Techniques And Its Application In Improving Imaging Quality Of Digital Cameras

Image restoration is of great interest in the digital image processing field. This paper studies a class of image restoration techniques including frequency domain algorithms and algebraic algorithms, together with space-invariant restoration algorithms and space-variant restoration algorithms. The inverse filtering restoration approach with polynomial approximation is particularly introduced. This approach is analyzed on its performance and used to correct field-curve photos from a low-end digital camera. Some techniques are used to solve the problems in this application. In addition, by combining adaptive restoration, energy-constraint, iteration with two-channel processing techniques, a scheme of energy constrained adaptive iterative restoration is proposed. It can also be used in solving the problem of space-variant restoration like the curvature of field.Firstly, the paper studies the image degradation model including the convolution model and the circulant matrix computing model. The classical restoration algorithms are described based on both the frequency domain frame and the linear algebraic frame. The frequency-domain restoration includes inverse filtering, Wiener filtering, power spectrum equalization and geometric mean filtering. And the linear algebraic methods includes non-constrained least square restoration and constrained least square restoration, some of which may lead to equivalent frequency domain solutions. Particular attention is given to the constrained adaptive restoration algorithm and its improved version that are derived from the linear algebraic frame. The coordinate transform restoration, a standard technique in space-variant restoration, is also introduced.Field-curve system is a typical space-variant degrading system. The inverse filtering with polynomial approximation is described in details, which can be used to restore the space-variant degraded signals such as field-curve images. Providing that the inverse transfer function of a degrading system is represented as the Taylor polynomial series, due to the fact that the product of the inverse transfer function and the Fourier transform of a degraded image is the Fourier transform of the corresponding restored image, and consequently, the image is restored by computing the linear combination of the original image and its derivative images. The restoration simulations of the one-dimensional and two-dimensional impulses and the analyses of performance argumentsprove the effectivity of this algorithm. By Simulating restoration of the images of different frequency components and computing the signal-noise-rate, the performance of this algorithm with different order approximation is understood.Energy consdrained adaptive iterative restoration algorithm is derived in this paper. Local energy constraint makes the algorithm automatically adapting the local information of images. Updating the constraining argument constantly during the Van Cittert iteration, which is used to solve the restoration equation, is of advantage in increasing the adaptability of the algorithm. The algorithm uses two-channel processing, which separates DC component of the observed signal, to remain the intensity of restored images and reduce the ringing. The algorithm is iterative and proceeds in space domain, so it can be applied in space-variant restoration like the field-curve. Meanwhile, increase of the computing due to the iteration is a problem too.The inverse filtering restoration approach with polynomial approximation is applied to correct the field-curve photos from an experimental digital camera. In order to depress the restoration noise, the scheme denoises the observed image, its 2-order derivative image, and the algebraic sum of derivative images. The denoising experiments and SNR analyse demonstrate that the three times of denoising is necessary. The experiment's results show that the restored images are not sensitive to the change of field-curve parameter. So choosing this parameter by experimenting is feasible. The estimation of field-curve p...