UBC Theses and Dissertations
Image interpolation with high visual quality Wang, Qing
Image interpolation or resizing has wide applications in areas such as digital photography, the printing industry, HDTV and computer graphics. The traditional methods for image interpolation are known to suffer from visual artifacts that degrade the subjective quality of images. The goal of this research is to study and develop new image interpolation techniques that provide images with improved visual quality. In our study, we focus on the two most harmful artifacts resulting from traditional image interpolation, zigzagging and blurring. In developing the new interpolation method, we concentrate on visually oriented interpolation techniques. We explore the feature-oriented and content-adaptive approach emerging from recent studies on image interpolation. To find out why and how zigzagging arises, we first study the isophotes, i.e., the equiintensity contours, of interpolated images. Based on this analysis, we design an interpolation scheme that employs interpolation grid that adapt to the orientation of image isophotes. This method yields much smoother isophotes in the interpolated images than the traditional interpolation methods. As a result, the zigzagging artifacts are largely suppressed. To remove the blurring effect, we propose a method of enhancing the contrast of expanded images. To do so, we first discuss the properties an edge-sharpening function should satisfy. From these we choose a family of edge-sharpening functions that have the desired properties. The proposed contrast enhancement method proves to be effective in removing the blur. We also show that evaluations based on the mean squared error (MSE) are not effective, thus we employ and improve the curvature-based measures to evaluate the performance of the proposed interpolation method. The experiments and analysis show that our proposed interpolation method is visually superior to traditional methods, and provide interpolated images with high visual quality.
Item Citations and Data