Multiresolution Segmentation Of Noise Image With Wavelet Transform

The principle problem of multiresolution segmentation of noise image issuppressing noise. Multiresolution analysis makes use of the independentinformation in different resolution. Mallat combines wavelet analysis withmultiresolution analysis, and multiresolution analysis based on wavelettransform becomes a focal field.Wavlet anaysis has been a new field in recent years. It has been accepted as avery important breakthrough of Fourier analysis, and has become a useful tool inimage and signal processing. The fundamental theory of wavelet transform isstudied in the paper, and then the application of mutiresolution analysis based onwavelet transform is explored in noise image segmentation. Through comparingbetween classical image segmentation algorithm and multiresolution analysisbased on wavelet transform, the excellence of multiresolution analysis can beembodied.Suppressing noise is the difficult problem in multiresolution segmentationbased on wavelet transform. To discern edge and noise in multiresolution edgedetection, the paper put forward a kind of adaptive threshold. Median filter is akind of nonlinear filter, and it has good abilities of keeping edge. A new kind ofadaptive fuzzy median filter is given in this paper, and this method can keepedge very well and reduce noise simultaneously. An image segmentationalgorithm based on median filtering and wavelet multiresolution is described inthis paper.The following is the image segmentation based on median filtering andwavelet multiresolution.1) Select wavelet basis used in edge detection.There are 3 rules to assure the quality of edge detection.Rule 1, The wavelet as edge detection filter should be high-pass filter;Rule 2, The wavelet basis should be consistent with the edge function inparity and symmetry. The edge of cell image to be detected in the paper is step edge, so the wavelet basis should be odd function; Rule 3, The wavelet used to detect image edge should be window function. It had better be compactly supported. So B spine orthogonal wavelet is chosen in the paper considering SNR and setting accuracy.2) The image is transformed by two-dimension wavelet, then the horizontal and vertical detail signal is acquired and then the module map and phase map is obtained based on them. Finally the edge map can be acquired after researching maximum wavelet transform value point and setting the lest points value zero.3) The edge map is filtered by adaptive median filter. The clutter can be removed under conditions of holding edge. The threshold selects the mean value of the statistics area.4) The ultimate edge is obtained after multiresolution detection.5) The ultimate edge links with edge tracing algorithm. The segmentation is complete now. Figure 1 is the image segmentation algorithm based on median filtering andwavelet transform. The local edge is more prominent after the image is transformed by waveletand the module value of the background and goal is little. The difference ofbackground and goal's grey value is big. So image can be segmented based on2-D grey-wavelet transform distribution. The Euclidean distance of background,edge, goal in 2-D grey-wavelet transform distribution is far, otherwise the...