| In the field of the digital image processing, the magnifying and the resolutionenhancement can be realized by image interpolation, which is very important practicallyin the sharpness processing and the 3-D displaying.In this paper, a new method of image size conversion is proposed based on 2-Dnon-separable linear phase finite-impulse response (FIR) interpolating filters andanti-aliasing filters. Comparing with the nearest neighbor interpolator, the bilinearinterpolator, the bi-cubic interpolator, the cubic B-spline interpolator, two-dimensionFFT interpolator and two-dimensional separable interpolating filters methods, themethod achieves better performance in the global error and local feature preservation ofimages, which is shown by the experimental results and theoretical analysis.Firstly, based on analyzing the basic theories of band-pass sampling, interpolationand decimation, effect on signal spectrum and fractional factor sampling, Thepoly-phase structure strategy, the multi-step structure strategy and the digital filterstructure strategy are exhibited as-realizing strategies for integral multiple decimationsystem. Secondly, the paper overviewed the requirements of the former interpolationalgorithms to the interpolation function. Then, several scaling algorithms of singleimage are realized, and also the results are compared. Finally, a new synthesis techniqueof 2-D non-separable linear phase FIR interpolating filters is put forward, which allowsto design polynomial-based interpolation filters with arbitrary frequency responses. Inaddition, designed filters characters are analyzed in this paper. The design of filter istranslated into an optimization problem with some constraints. Especially, regularity asone of the constraints assure the flatness in passband and the high attenuation instopband.A lot of experiments in this paper validate that the two methods can achievepreferable results to some image data such as natural scenery, animal pictures andpersonality pictures. |
PDF Full Text Request