A Bidimensional Local Mean Decomposition Algorithm

The multiscale analysis method is able to decompose an image into a set of different frequency scale images that provide much information for further image processing.In this paper,a new bidimensional LMD(NBLMD)is proposed to extract feature information at multiple scales.Firstly,morphological reconstruction or a fast eight-neighbor window is developed to detect local extremas of images.A method based on Delaunay triangulation is proposed to determine the adjacent extreme points,and then the local mean function and local envelope function are calculated effectively.Thirdly,a new stopping criteria of 2D-sifting process is proposed based on the position information of bidimensional product function(BPF)and the value of the envelope estimation function.Finally,the BPFs are extracted effectively in a sifting process.The experimental results show that NBLMD is capable of implementing multiscale analysis of images effectively and quickly.The comparison results illustrate that NBLMD outperforms bidimensional LMD(BLMD)and bidimensional empirical mode decomposition(BEMD)for image multiscale analysis.