Curvelet, as a revolutionary method of Multiscale Geometric Analysis, is better suitable in analyzing the property of 2D curves and straight line like edge in comparison to the conventional wavelet method. It is revolutionary also in the sense of higher accuracy close to limits and better sparse representations. This is more effectively demonstrated in the second generation curvelet transform. Understanding plenty work wants to do, as the study of curvelet transform has just started.This paper introduces the common transforms followed by a discussion of our second generation curvelet transform, the principle and the way of calculation. We applied curvelet transform into image fusion: the source images are decomposed using curvelet transform first and combine coefficients on the corresponding scale using Fusion Rules. Finally we get fusion result by image reconstruction. Suitable fusion rules are selected by experiments. Experiments demonstrated that all of curvelet transform calculations are more effective than wavelet transform. In particular, fusion rules based on the edges is an especially efficient result.Targeting the shortcomings of conventional method in denoising, this paper develops well the Threshold Function in its procedure and introduces S Function in common use at neuron network. We propose a novel method in minimizing noise based on the second generation curvelet transform, this method extracts the edge information in source images after denoise, then the image obtained from curvelet threshold denoise, and finally use processed edge information and directly denoise image fuse with the fusion rule. It recovers the characteristic in the original image much better. |