Edge-Directed Interpolation Based On Statistical Estimation

Image is the main resource by which human can access and exchange information.Therefore, Image processing applications will inevitably involve human social life andwork mostly. Image interpolation and enlarging technology is an important component partof image processing, and plays a very important role. The difficulty of image interpolationlies in how to balance between keeping the smooth region of the background and theintegrity and continuity of the edge, so as to reduce image distortion.In this paper we firstly introduce some traditional image interpolation algorithms,including the nearest neighbor interpolation，linear interpolation and some other nonlinearalgorithms. And then we sort and classify the latest algorithms, including edge-basedinterpolation, region-based interpolation and some other algorithms, such as fractalinterpolation, wavelet interpolation and PDE-based interpolation.Secondly, we expound the basic features of image edge and some classic edgedetection method. In image interpolation techniques, an effective method should make aspecial treatment on edge pixels because it generally carries the most information ofimage. So we discuss the Canny edge detector in detail as well as some simulation resultsand related analysis. It turns out that the canny detector operator can obtain a betterbalance between noise suppression and edge extraction. Besides, the edge width is onepixel width, which is consistent with the requirements of our algorithm.Finally, the image random linear model is introduced, and the classic method basedon edge-directed namely NEDI is improved. We estimate a high resolution edge map byCanny detector to get more accurate location and direction for the edge pixels, and thenclassify the pixels to be interpolated to three categories including edge area, homogeneousarea and transition area. The unknown pixels are calculated based on different formulas, especially for transition area, the classical Grubbs rule is introduced to make asample-selection before interpolation. Experimental results show that our algorithmperforms better in visual and higher objective quality than NEDI, iNEDI when applied tohigh-contrast image.