فصلنامه پردازش علائم و داده ها
سال نهم شماره 1 (پیاپی 17، بهار و تابستان 1391)

Automatic capture and analysis of human motion، based on images or video is important issue in computer vision due to the vast number of applications in animation، surveillance، biomechanics، Human Computer Interaction، entertainment and game industry. In these applications، it is clear that 3D human pose estimation is an essential part. Therefore، its accuracy has a great effect on the performance of these applications. Because of the variation in appearance and articulations of human، self-occlusion and high dimensional state-space of human pose، 3D human pose estimation from image observations is a challenging problem. In this paper، a new method of 3D human pose estimation from multi-view video sequence is introduced. In the proposed method، instead of seeking directly over the high dimensional states-space of human pose and employing the complex inferring algorithms، a hierarchical search method with distinct objective function for each part of the body and direct optimization methods is employed. Advantages of the proposed method are: automatic initialization، labeling of parts of the body contour and using separate objective function for different parts of the body. Experimental results demonstrate that the proposed method can be effectively used as a marker-less system to estimate 3D human pose in a multi-view sequence.

In this paper، we show that the problem of grammar induction could be modeled as a combination of several model selection problems. We use the infinite generalization of a Bayesian model of cognition to solve each model selection problem in our grammar induction model. This Bayesian model is capable of solving model selection problems، consistent with human cognition. We also show that using the notion of history-based grammars will increase the number and decrease the complexity of model selection problems in our grammar induction model. This results in the induction of a better grammar which leads to 9. 1 points increase in F1 measure، for parsing the section 22 of Penn treebank in comparison with a similar model that does not use history-based grammar induction techniques.

Performance of speech recognition systems is greatly reduced when speech corrupted by noise. One common method for robust speech recognition systems is missing feature methods. In this way، the components in time - frequency representation of signal (Spectrogram) that present low signal to noise ratio (SNR)، are tagged as missing and deleted then replaced by remained components and statistical information of clean speech. In this article a new approach of missing features method based on compensation are proposed. A Bidirectional Neural Network (BNN) was developed and implemented in order to modify unreliable components in input feature vectors and improve the overall recognition accuracy. Distorted components in feature vectors were estimated in accordance with the latent knowledge in the hidden layer of the neural network. This knowledge is obtained by training with clean and noisy speech، simultaneously and is mostly induced by reliable and less influenced components by the irrelevant variations in speech signal. In this approach، there is no need to identify missing components that is a challenging issue in the field of robust speech recognition based on missing feature method because reconstruction is done on all components (whether reliable or unreliable)، in order to become more similar to the clean speech component. This point is a very significant advantage that has been achieved in this article. Comparing the results of these two methods shows that using Missing feature methods، 4. 2% improvement were obtained in the accuracy of speech recognition for noisy signal by SNR=0dB، whereas improvement value increased to 8. 5%، using bidirectional neural network for the same signal to noise ratio.

In this paper a new method for the online recognition of handwritten Persian characters has been proposed which uses a set of simple features and Support Vector Machine (SVM) as a classifier. The task of preprocessing allows us to equalize feature vectors from different characters. This algorithm is implemented in two steps. In the first step، input character is classified into one of eighteen groups of main strokes of characters and in the second step، position، number، and the shape of sub-strokes determine character type. For example to recognize the character ‘ت’، in the first step the character will be classified to group of letters ‘ب، پ، ت، ث’ based on main stroke shape and then classification is done using information of the sub-strokes. In the final step، post processing، we rectify previous step results employing unmatched conditions between main stroke and sub-strokes. Consider a main stroke «ل» with a point at the top of that in this situation post processing step will change result to letter «ن». The experimental results -which is based on Online-TMU database- show that the recognition rate of the main strokes of the characters is 94% which reaches to 98% using the information of sub-strokes.

Military Information Extraction techniques are interested for military managers and commanders. But usual information extraction techniques cannot be used for that domain، because military corpus has special structure that differs from non-military corpus. In this paper the military documents structure is compared with non-military documents structure. Moreover a new classification is proposed for military documents. IE system effects are also surveyed for each class. Then a new composite approach is proposed according to the experimental results. The new approach is based on Wrapper Induction and CRF Model. Test bed is also prepared from battlefield reports to evaluate the proposed system. The improvement is shown based on tow metrics namely Recall and F-Measure. The usefulness of the proposed system is also demonstrates.