Image Interpolation Based On The DCT Domain

With the continuous development of computer and communication technologies,a variety of different resolution display devices have come into being, an endlessstream of applications requiring high resolution. Image interpolation is one of themost effective techniques to satisfy these requirements, these needs to better reflectthe importance and urgency of the image interpolation. Image interpolation is themost basic and important research topic in the fields of image processing recentyears, many image interpolation algorithms have been proposed. However, a largenumber of image interpolation algorithms focus on the spatial pixel value domainwhile rarely work in the frequency domain. Therefore, this article focuses on thefrequency domain image interpolation.With H.264/AVC, AVS video coding standards widely used in compressedvideo, a new generation of video coding standard HEVC being developed, whichmakes the frequency domain interpolation algorithm also has a wide range ofscenarios, that is, if in the decoding of the video coding system side we get thereconstructed DCT coefficients, the frequency domain interpolation algorithmdesign can be applied directly.In this paper, we design a frequency-domain interpolation algorithm for thereconstructed DCT coefficients block, compared with traditional spatial domaininterpolation algorithm, the computational complexity of the compressed domainimage interpolation method is less than that of the class of adaptive spatial domaininterpolation methods, the interpolation in the compressed domain is much fasterthan the interpolation directly in the spatial domain. According to our derivedformula and the statistical properties between the spatial domain and the frequencydomain interpolation consistency, the corresponding relations of the spatial domianinterpolation and the frequency-domain interpolation can be established.Then weuse the Wiener filtering in the frequency domain interpolation to train out a set ofoptimal interpolation coefficients based on the training set which can improve theeffect of image interpolation. The experimental results show that our proposedalgorithm compared with the traditional interpolation methods has better subjectiveand objective effects. In addition, we improve the proposed frequency-domaininterpolation algorithm, the specific performance by adding the classification of theWiener filtering weights training, different image categories having differentinterpolation filters for image interpolation. Besides, we also tried interpolationmethods on the reorganization of the DCT coefficients or changing the modeselection in order to achieve better interpolation effect. It is noteworthy that our proposed frequency domain interpolation algorithmcan be done parallel to the block level that can quickly and effectively enhance theeffect of the interpolated images which can be used directly for the video codingsystem. Therefore, as a frequency-domain interpolation algorithm, this algorithmhas a theoretical basis and practical applications.