Earticle

Tracking with a discriminative classifier becomes popular recently. The online updating makes it easy to adapt to target appearance variations. However, this also brings drifting problem. It’s necessary to find a tracking method with strong adaptivity and anti-drifting ability. In this paper, an online semi-supervised boosting method is proposed at first, and based on it, we propose a novel tracking framework that treats samples differently when updating the classifier under different conditions. This tracking framework can significantly alleviate the drifting problem and keep adaptive enough to appearance variations. Experimental results on challenging videos show that our method can track accurately and robustly， and outperform many other state-of-the-art trackers.

A hybrid microcalcification detection method based on Fisher discriminant and SVM was presented because signal-to-noise ratio of mammogram image was very low, and microcalcifications were very small and their shape was irregular. Firstly, low frequency information of tissue was removed by wavelet transform in order to reduce the tissue effect to microcalcification segment. Secondly, Fisher discriminant was adopted to find optimum threshold, meanwhile microcalcification was segmented. Lastly, SVM classifier was adopted to recognize true microcalcifications. Experiment results showed that Fisher discriminant could validly segment microcalcifications and the number of false positive targets was less than OSTU’s. Detection ratio of our algorithm was about 97%.

Probabilistic latent semantic analysis (pLSA) has been widely used by researchers for human action recognition from video sequences. However, one of the major disadvantages of pLSA and its other extensions is that category labels of training samples are not fully used in model learning procedure for classification task. In this paper, a supervised pLSA (spLSA) model is proposed for overcoming this drawback. By adding an observable category variable to generative process of classic pLSA, spLSA is endowed with more discriminative power. Thus, this model provides a unified framework for semantic analysis and object classification, where the topics formulation is guided by spLSA towards more discriminative and the mapping between the topics and the action categories are described in a fully probabilistic manner. Experimental results show that spLSA substantially outperforms pLSA and achieves comparable or better performances than latent dirichlet allocation based supervised models and other state-of-the-art methods.

This paper presents signal decimation and interpolation techniques under a multiresolution frame work for both lower and higher dimensional applications. New classes of non-linear basis functions have been derived from the sigmoid activation function extensively used in artificial neural networks (ANN). It has been shown that the proposed non-linear basis functions are well suited for interpolation/approximation of band limited signals. An efficient scheme for band limited signal interpolation has been introduced. Fast IIR digital filters (inverse filters) have been derived from the combinatorial theory in connection with the proposed basis functions. The proposed inverse filters can easily be implemented recursively with three multiplications and additions only. Further, the factorization of higher order filters for easy implementation has also been considered. Frequency response characteristics for the pre-filters and their corresponding interpolators are presented to reveal the quality of interpolation. An experiment has been carried out to interpolate a discrete sequence of length 33 into a sequence of length 257 (with a zooming factor of 8). Second part of the paper presents another efficient scheme for image decimation and interpolation. Experimental results on image data compression have been presented to justify the use of the proposed technique.

People today spend longer seated resulting from changes in demand on the workforce. As a result there is a need for a greater understanding of factors affecting pressure sore formation and comfort in general. In order to monitor the body-cushion interface temperature, we have developed a portable five-channel temperature measuring system which can be powered by a laptop. An Empirical Mode Decomposition (EMD) was used to remove noise of thermal data between body and seat contact surface. The performance of this data driven filter was compared with three other filters (medium filter, adaptive filter and wavelet filter) with the help of the goodness of fitness statistics as judgment criteria. Results showed the EMD-based filter worked better than traditional de-noising algorithms with the lowest RMSE (root-mean-square-error) and the highest R2 values.

On the basis of the flow feature in the micro blood vessels, a method to track the trace of cell movement was proposed in this paper. The algorithm employed the space domain enhancement, adaptive threshold segmentation techniques. For the sake of calculating blood flow, the active contour-based template matching method was applied to sequential images. The experimental results show that this method can accurately track movement of tiny particles and estimate its moving speed in the micro-blood vessels.

This work presents an approach for the analysis of abnormality in the cervical cells based on Texture and presence of Hyperchromasia, which are two important morphological features based on which one can distinguish between normal and abnormal cervical cells. The proposed approach is implemented in MATLAB®, a high level, interactive environment for data visualization/analysis/computation. This may help pathologist in identification of cervical cancer from Pap smear images and help in early diagnosis.

Skull stripping head magnetic resonance (MR) images is a classification problem, which is a difficult task due to the convoluted brain surface. This paper presents an automated, robust, and accurate brain volume extraction method (BVE)based on a combined image segmentation and bias estimation method (ISBEM) and a 3D mathematical morphology method. The ISBEM segments brain tissues efficiently and accurately, while estimating the bias field. The new 3D mathematical morphology method can remove the noncerebral tissues efficiently. The results show that our method can obtain more accurate results.

In allusion to remove Racian noise while lessen the loss of details as low as possible, this paper proposed an filter algorithm which comprehensive utilize Multi-Objective Genetic Algorithm (MOGA) and Shearlet transform based on a Multi-scale Geometric Analysis (MGA) theory. First, it performs a wavelet multi-scale decomposition of image. Then, it builds target function in MOGA by several evaluation methods such as Signal to Noise Ratio (SNR). Third, it uses the MOGA to optimal coefficients of Shearlet wavelet threshold value in different scale and different orientation. Finally, it obtains the composite image by using inverse lifting wavelet transform. Experimental results show tha our proposed new algorithm presented here is more effective in removing Rician noise, and giving better Peak Signal Noise Ratio (PSNR) gains, without manual intervention in comparison with other traditional filters.