Image Interpolation Algorithm Based On The Two-dimensional Empirical Mode Decomposition

Empirical Mode Decomposition (EMD) can adaptively decompose non-linear and non-stationary signal corresponding to different signal frequency. For this reason, it puts forward a new direction for future signal processing.EMD, which successfully applied in one-dimension signal processing, many experts and scholars extend it to2-Dimensional data, what is named Bidimensional Empirical Mode Decomposition, BEMD. It obtained good effect in seismic, biomedical, image feature analysis etc. BEMD can decompose image to some features (Intrinsic Mode Functions, IMF) with different frequency by a sifting process. The sifting process is realized using neighboring window to detect regional maxima (also minima) and radial basis function for surface interpolation.After BEMD process, the first IMF shows image feature distinct, so we can detect the direction of image feature successfully. Then, we propose a new image interpolation method. First, we process image based on BEMD and get the first IMF, also it is the IMF with highest frequency. Over an amount of experiments have been made, we get the algorithm suit for detect the feature direction and interpolation relatively. For the lower frequency features, we use bilinear to interpolation. Finally, we obtain the magnified image which by adding every features after interpolation.Use the theory above, through experiment prove, the algorithm we use can reduce the Mean Square Error (MSE) by a large margin compare with traditional method. Furthermore, the edge of the feature, which is sensitive for the vision, our approach can eliminate the ladder effect as more as possible, and then restore the image as possible as desired.