Image-based Data Mining Method

The developing trend of the Data Mining is diverse, which the flexibility of the technic of the data mining is known clearly. Since the improvement of the era of the information , more technic is widely interinfiltrated which involved the data mining. In this paper, the technic of the data mining is used in the field of the image processing. Lots of the researching work has been done , which combine the technic of the data mining to the technic of the image processing, to solve the problem in remote sense image processing. The model of the data mining based on image has been built and every step in the model has been analyzed theoretically and applied in reality. The main contents of this thesis are summarized as follow:(1)The course of the data mining development is reviewed and the role which the data mining played in KDD is definite. The components has been classified according to the task of the data mining, and in each component the technic has been summarized. The model which apply to image data mining is proposed based on these components. (2)The image data reduction is very important in remote sense image processing, two methods in data reduction proposed in this paper, which one is based on wavelet theory and the other is based on clustering analysis. The method based on wavelet theory is discussed in the respect of the relation between the wavelet filter and the linear filter. The 9/7 wavelet lifting algorithm, which is the Biorthogonal wavelet has the linear phasic, is applied in the reduction and has a good result. The method based on clustering analysis put the image into subimage by the quarter tree method according to the local characteristic, and in the subimage finishing clustering. Based on the center of the subimage clustering we do the clustering to get the clustering on the whole image. (3)The distribution and the direction in texture is discussed. The texture segmentation method which based on the self-adapted texture windows selected is proposed. The Gabor filter is used to extract the characteristics. In this method, the jet which is a kind of the multi-scaled analysis methods is used, and get a good result. GMRF is a kind of segmentation methods which combine gray characteristic and texture characteristic as its characteristic. GMRF segments image effect. In this method, to combine the MRF and the characteristic whichhas the Guass distribution is the key , and we discussed in the paper. (4) Building the image characteristic database, visualization analysismethod is used on the image characteristic to get some usefulinformation.