Analysis of Interpolation Methods in the Image Reconstruction Tasks

Abstract

The article studies the interpolation methods used for image reconstruction. These methods were also implemented and tested with several images to estimate their effectiveness. The considered interpolation methods are a nearest-neighbor method, linear method, a cubic B-spline method, a cubic convolution method, and a Lanczos method. For each method were presented an interpolation kernel (interpolation function) and a frequency response (Fourier transform). As a result of the experiment, the following conclusions were drawn: - the nearest neighbor algorithm is very simple and often used. With using this method, the reconstructed images contain artifacts (blurring and haloing); - the linear method is quickly and easily performed. It also reduces some visual distortion caused by changing image size. Despite the advantages using this method causes a large amount of interpolation artifacts, such as blurring and haloing; - cubic B-spline method provides smoothness of reconstructed images and eliminates apparent ramp phenomenon. But in the interpolation process a low-pass filter is used, and a high frequency component is suppressed. This will lead to fuzzy edge and false artificial traces; - cubic convolution method offers less distortion interpolation. But its algorithm is more complicated and more execution time is required as compared to the nearest-neighbor method and the linear method; - using the Lanczos method allows us to achieve a high-definition image. In spite of the great advantage the method requires more execution time as compared to the other methods of interpolation. The result obtained not only shows a comparison of the considered interpolation methods for various aspects, but also enables users to select an appropriate interpolation method for their applications. It is advisable to study further the existing methods and develop new ones using a number of methods considered as a basis.