Adaptive Digital Image Interpolation And Its Application

Image interpolation is one of the key technologies in image/video processing. It is fundamental to many digital image operations, such as scaling, rotation, and geometric correction. This thesis studies a class of traditional image interpolation techniques, including spatial domain algorithms and space-frequency domain algorithms, together with space-invariant algorithms and space-variant algorithms. After analyzing these algorithms, we particularly introduce an image interpolation deblurring operator based on two-dimensional rational filter. We analyze its performances and propose a fast and effective image interpolation scheme by combining this operator with space-invariant interpolation technique, two-channel signal processing technique, iteration algorithm and contrast constraint. We use it to improve resolution of images acquired by a low-end digital camera. According to the traits of the special cameras, some application problems are solved. Image degradation and restructuction model is built firstly according to actual image acquiring system. Shannon information theory, the base of signal sampling and interpolation, is described in detail. The ill-conditioning property of super-resolution problem based on single observed image is discussed at large. How much can interpolation techniques do in image super-resolution is ascertained by followed analyses. In the frame of Shannon sampling theory, we introduce linear space-invariant interpolation techniques in spatial domain and frequency domain. Frequency algorithms include spectrum zero padding and inserting zero-value samples followed by a frequency domain ideal low-pass filter. Spatial algorithms including nearest neighbor interpolation, linear interpolation, cubic convolutional interpolation, polynomial interpolation, spline interpolation and Gaussian interpolation. We analyze traits of zoomed images produced by these linear space-invariant interpolation techniques, and find their common defect: lacking of high frequency components in the detail area and failing to preserve the smoothness of background and steepness of edges at the same time. Some novel adaptive image interpolation techniques, such as two-channel interpolation, fractal interpolation, edge-directed interpolation, partial differential equation interpolation and rational interpolation, are introduced in detail and their advantages and disadvantages are discussed. Adaptive interpolation techniques produce more vivid zoomed images compared with traditional interpolation techniques, but they are difficult to be used in actual image processing system for their huge computational burden. Interpolated image pos-processing techniques, which try to enhance the images processed by space-invariant interpolation algorithms by using some deblurring operators, have more advantage in actual application. Rational filters have strong edge preserving capability because their filter coefficients are controlled by the signals they will process. This property is useful to deblurring of interpolated images. We proposed an adaptive deblurring operator based on high-pass rational filter. Its two-dimensional form is deduced for image processing. The simulation experiments on a step signal prove the edge rebuilding capability of this operator. We propose a fast and effective image interpolation scheme based on this rational deblurring operator. In this scheme, we use a contrast constraint to ensure interpolated images free from over-steep artifacts induced by too much iteration. Because the background of zoomed image needn't be enhanced, a two-channel processing idea is combined in this scheme. The computational burden of scheme is reduced greatly. The performances of this scheme are evaluated by visual comparison of images and their Fourier spectrum, and then by some objective measures, such as image resolution, peak signal noise ratio and Q value. This adaptive image interpolation scheme is applied in resolution enhancement of images obtained by an experimental camera. The problem of color image processing is solved and computational burden is reduced more. The camera, which integrates the scheme into its background software, will get higher performance cost ratio.