Wavelet-based Multiscale Image Processing Method And Its Applied Study In Pulp Fibre Detection

Wavelet analysis, since its breakthroughs in 1980s, has proven to be a valuable tool in many application fields and has been applied in almost all information science fields. As a tool of time-frequency analysis after Fourier analysis, its application covered the field of image processing, such as image pre-processing, image compression and image retrieval, feature extraction and pattern recognition. Grounded on the research of the image processing and characteristics extraction of pulp fibre, this paper discussed the following five problems in detail on the basis of the multiscale representation of image.1. The directional characteristic of wavelet transform. Multiscale and multidirection are natural attributes of image, which can help to describe the image more objectively, hence more essences of image can be revealed. Although wavelet transform can depict image in multiscale way freely, it cannot describe the directional characteristic of image effectively. So we have designed a directional filters based on wavelet transform, which can represent image in any direction. Then image can be represented in multiscale and multidirectional way according to the image characteristics or our demands. We also have used the directional filters in edge detection of pulp fibre image and achieved good results.2. The fuzzy multiscale filter of wavelet. There are two problems when wavelet is used in multiscale decomposition. One is that the finer scale indicates the local detail information of edge, localization being precise but noise predominant. The other is that the coarse scale indicates the global information of image, noise distribution being lower but localization vague. To detect edge and eliminate noise effectively at the same time, we consider to synthesize the multiscale information of image through some rules. The edge points can be extracted by taking advantage of finer and coarse scale respectively. We put forward a fuzzy multiscale edge detection method, that is, first, synthesize scale information via fuzzy rules;second, transform the image information from wavelet domain into fuzzy space;third, use the fuzzy decision method to synthesize the fuzzy subsets, and at last extract the edge information.3. Image denoising based on statistical model in wavelet domain. Theeffectiveness and objectivity of image model will directly affect the precision of denoising. We attempted two different statistical models to study the denoising characteristics. Firstly, the wavelet coefficients are modeled as generalized Gaussian distribution, and the self-adaptive strategy is studied on the basis of maximum a posteriori probability rule. This method is designed according to the respective pixel points. Secondly, the wavelet coefficients are modeled as mixed Gaussian distribution. For the correlation among the three subbands (horizontal direction, vertical direction and diagonal direction), a certain vector is used to describe the subbands model. The hidden Markov model is used to describe the correlations between the father node and son node. In addition, Bayes estimator is also used in denoising.4. The corner detection algorithm. Because corner can extract the important characteristics of objects while effectively reducing data, it is widely used in shape detection and image match. The traditional corner detection method has some defects, such as the appointed local support region size and the complicated curvature method. A slide window method is put forward on the basis of covariance matrix of boundary curve. The method can not only adjust the local support region according to the bending index, but also show the bending characteristics flexibility. The eigenvector of covariance matrix is used to denote the curvature direction. The two-dimension image is converted into one-dimension characteristic curve. Wavelet transform is the tool to extract the singularities of one-dimension, so the corner of two-dimension image is extracted effectively. The new corner detection algorithm is used to extract the shape characteristics of pulp fibre and the results show the effectiveness.5. The effective detection of pulp fibre. The detection of paper fibre is significant. We applied the theory and technology of image processing in it and put forward a new approach in this field. The experiment platform is designed to acquire the dynamic on-line fibre image, which overcomes the shortcoming of off-line fibre measurement. An effective judge criterion of paper fibre characteristics is put forward. The multiscale image processing algorithm is designed to extract the shape parameters according to the physical characteristics of fibre. Experiment results indicate the reasonableness and effectiveness of our method.