可视跟踪就是利用图像处理、模式识别的方法发现视频序列中与指定目标图像最相似的部分，在兼顾实时性的基础上提高跟踪算法的稳健性一直是可视跟踪研究中的前沿和热点。本文提出利用目标历史模型和当前匹配位置处得到的观测模型对目标核函数直方图进行Kalman滤波，从而对模型进行及时更新。首次提出把滤波残差作为样本进行假设检验，将其结果作为模型是否需要更新的准则。论证了Mean-shift框架下跟踪变尺度目标的充分条件，提出了“后向跟踪－形心配准”的核窗宽自动选取算法。实验验证了所提方法的有效性。

A software tool for data mining (DMiner-I) is introduced, which integrates pattern recognition (PCA, Fisher, clustering, HyperEnvelop, regression), artificial intelligence (knowledge representation, decision trees), statistical learning (rough set, support vector machine), computational intelligence (neural network, genetic algorithm, fuzzy systems). It consists of nine function models: pattern recognition, decision trees, association rule, fuzzy rule, neural network, genetic algorithm, HyperEnvelop, support vector machine, visualization. The principle, algorithms and knowledge representation of some function models of data mining are described. Nonmontony in data mining is dealt with by concept hierarchy and layered mining. The software tool of data mining is realized by Visual C++ under Windows 2000. The software tool of data mining has been satisfactorily applied in the prediction of regularities of the formation of ternary intermetallic compounds in alloy systems, and diagnosis of brain glioma.

Introduction: With the promotion and application of digital imaging technology in the medical domain, the amount of medical images has grown rapidly. However, the commonly used compression methods cannot acquire satisfying results. Methods: In this paper, according to the existed and stated experiments and conclusions, the lifting step approach is used for wavelet decomposition. The physical and anatomic structure of human vision is combined and the contrast sensitivity function (CSF) is introduced as the main research issue in human vision system (HVS), and then the main designing points of HVS model are presented. On the basis of multi-resolution analyses of wavelet transform, the paper applies HVS including the CSF characteristics to the inner correlation-removed transform and quantization in image and proposes a new HVS-based medical image compression model. Results: The experiments are done on the medical images including computed tomography (CT) and magnetic resonance imaging (MRI). At the same bit rate, the performance of SPIHT, with respect to the PSNR metric, is significantly higher than that of our algorithm. But the visual quality of the SPIHT-compressed image is roughly the same as that of the image compressed with our approach. Our algorithm obtains the same visual quality at lower bit rates and the coding/decoding time is less than that of SPIHT. Conclusions: The results show that under common objective conditions, our compression algorithm can achieve better subjective visual quality, and performs better than that of SPIHT in the aspects of compression ratios and coding/decoding time.

Automatic identification of landmarks in cephalometry is very important and useful for orthognathic surgery. In this paper, a computerized automatic cephalometric analysis system (CACAS) based on image processing is presented. For original X-ray image, median filtering and histogram equalization is used to improve image quality. The edge of a X-ray image is detected by wavelet transform and Canny filter. Seventeen landmarks in cephalometry are successfully identified by knowledge-based edge tracing and changeable template. Seventy three measurements based on distances, angles, ratios among landmarks are computed automatically. The reliability of landmarks and the validity of measurements are compared between automatic and manual operation. The values of measurements by CACAS are preciser and more reliable, the mean error for linear measurements is less than 0.9mm; the mean error for angular measurements is less than 1.2dg. The rate of validity is over 80% even if the image quality is poor. For an image with high signal to noise ratio, the rate of validity of landmarking and measurements by the CACAS system is over 90%.

Fast computation of Zernike moments from normalized geometric moments has been developed in this work. The computation is multiplications free and only additions are needed to generate Zernike moments. Geometric moments have been generated using Hatamian's filter up to high orders by a very simple and straightforward computation scheme. Other kinds of moments (i.e., Legendre, pseudo Zernike. etc.) can be computed using the same algorithm after giving the proper transformations that state their relation to geometric moments. Proper normalizations to geometric moments are necessary so that it can be used in the efficient computation of Zernike moments. To ensure fair comparisons, recursive algorithms are used to generate Zernike polynomials and any other coefficients. Computational complexity model and the test programs show that the speed up factor of the proposed algorithm is superior with respect to other fast and/or direct computations. To our knowledge, it is the first time that Zernike moments can be computed in real time rates which encourages the use of Zernike moments features in different image retrieval systems that supports huge database such as XM experimental model stated for MPEG-7 experimental core. It is concluded that choosing direct computation would be impractical.