| Wavelet analysis is a tool of time-frequency analysis after Fourier analysis. In the field of image processing, its application covered imaging technique, image pre-processing, image compression and transferring, image registration, image analysis, feature extraction and pattern classification, etc. In this paper, it's researched on wavelets application in the fields of multi-scale edge detection, remote sensing image processing and medical imaging.The traditional methods of edge detection are based on one-order derivative's maximum, or two-order derivative's zero-crossing. This kind of edge definition is very sensitive to noises. And thus, edge detection should be carried out in large scale, by which the image was smoothed. One of the shortcomings of edge detection in large scale is that it's difficult to locate edge precisely, which will make mistakes in pattern recognition based on edge features. With multi-scale characterization, wavelet analysis was widely used to multi-scale edge detection. In this paper, it was proved that, wavelet-based multi-scale edge detection would keep edge positions very well, if symmetric bases were used in wavelet transform. Furthermore, an algorithm of multi-scale edge detection based on bi-orthogonal symmetric wavelet was put forward, with which, "good edges" will be obtained while the edge positions will be kept well.High-frequency information after wavelet transform is a kind of one-time feature which is not convenient for pattern classification. By statistics or other kind of processing, two-time features can be obtained. With wavelet based gray-gradient symbiosis matrix to compute the ununiformities of gray distribution and gradient distribution, a kind of pollution index was defined. Then all of these features were used to Taihu Lake water pollution analysis. And a kind of contour map representing pollution degree was obtained, by which it can be known about the pollution situation of whole water area.After image registration and some pre-processing, the 0-level data of satellite remote sensing image still contains some noises such as strip noise. The width of strip noise is relative to the number of detect-units in satellite detector; and the intensity of strip noises is relative to areas on the earth surface. According to the width of strip noises, the frequency band corresponding to the strip noises can be figured out. Thus, after wavelet transform, by setting proper wavelet coefficients zero, the strip noises can be removed effectively. Ideal result has been obtained in the practical processingto 0-level data of Cbers-1.It can increase information quantity, improve vision effection, and therefore advance the explanation ability of one frame of image, by carrying out data fusion with remote sensing images got from different satellites or with images of different bands got from same satellite. After proper wavelet transforms, by carrying out pixel-level fusion for images with same size, using sub-images with same transform hierarchical, or for images with different size, using sub-images with same or different transform hierarchical, the image can both be enhanced. In this paper, two kind of wavelet based fusion algorithm of remote sensing images are improved. By experiments on TM images, the improved algorithms are both better than original ones according to the Shanon entropy and mean gradient.The advance of optical technique and pattern recognition technology has promoted development of compute-aided diagnosis, medical imaging and post-processing of medicine image. Combining wavelet features and morphology features of cancer cells in shape and structure, a cancer-cell auto-recognition method is put forward. The experiments on hybrid examples of lymphocyte, meothelial and adenocarcinoma shows that the recognition ratio using this method is nearly 100%.Magnetic resonance imaging (MRI) is a noninvasive method of imaging. The technique of tagged MRI provided most convenience for cardiac motion tracing and its reconstruction. The noises in MR images...
|
PDF Full Text Request