SIGNAL AND DATA PROCESSING
Year:2016 | Volume:- | Issue:1 (SERIAL 27)|Papers Count:10

In this paper, a new structure for image encryption using recursive cellular automata is presented. The image encryption contains three recursive cellular automata in three steps, individually. At the first step, the image is blocked and the pixels are substituted by reversible cellular automata. In the next step, pixels are scrambled by the second cellular automata and at the last step, the blocks are attached together and the pixels are substituted by the third cellular automata. Due to reversibility of cellular automata, the decryption of the image is possible by doing the steps reversely. The experimental results show that encrypted image is not understandable visually, also this algorithm has satisfactory performance of quantitative assessment from some other schemes.

In this paper an efficient approach based on the Extreme Learning machine, for identifying and modifying image pixels which are contaminated by salt pepper is presented. The proposed algorithm uses an extreme learning machine to identify noisy pixels in digital image. Decision of the ELM is based on six inputs including: central pixel, ROAD factor and four measurements of SD-ROM filter. In the next phase, best values for each of the noisy pixels is estimated using an adaptive median filter. The results of the performance evaluation of classification, represents the high capability of input features in discriminating noisy pixels from noise-free pixels. To evaluate the performance, the output image of the proposed algorithm is compared with the output of several other common filter in terms of PSNR ratio. The numerical results obtained from the tests indicate efficiency of proposed filter in terms of qualitative and quantitative criteria.

The automatic detection of words that have semantic roles in sentences, and determining the type of their semantic roles (i.e. agent, patient, source, ...) may improve the quality of systems based on natural language processing tasks such as information extraction, question answering, summarization, and machine translation. In this task, which is called semantic role labeling or shallow semantic parsing, syntactic parsers are usually employed for extracting syntactic features and the type of syntactic representations that the parsers use affect the overall precision of the semantic role labeler. In this research we present a machine-learning-based semantic role labeler for Persian sentences in which syntactic features are extracted by a dependency parser. In many aspects, the presented system performs better than previous semantic role labelers for Persian, which all used shallow syntactic parsing.

In this article, growable deep modular neural networks for continuous speech recognition are introduced. These networks can be grown to implement the spatio-temporal information of the frame sequences at their input layer as well as their labels at the output layer at the same time. The trained neural network with such double spatio-temporal association structure can learn the phonetic sequence subspace. Therefore, it can filter out invalid phonetic sequences in its own structure and output valid sequences. To evaluate the performance of these growable neural networks, we used FARSDAT and BIG FARSDAT datasets. Experimental results on FARSDAT show that deep modular neural networks outperform the phone accuracy rate of GMM-HMM models with an absolute improvement of 2.7%. Moreover, developing deep modular neural networks to a double spatio-temporal association structure improves their result by 5.1%. As there is no phonetic labeling for BIG FARSDAT, a semi-supervised learning algorithm is proposed to fine-tune the neural network with double spatio-temporal structure on this dataset, which achieves a comparable result with HMMs.

The performance of automatic speech recognition (ASR) systems is adversely affected by the variations in speakers, audio channels and environmental conditions. Making these systems robust to these variations is still a big challenge. One of the main sources of variations in the speakers is the differences between their Vocal tract length (VTL). Vocal tract length normalization (VTLN) is an effective method introduced to cope with this variation. In this method, the speech spectrum of each speaker is frequency warped according to a specific warping factor of that speaker. In this paper, we first developed the common search-based method to obtain the appropriate warping factor over a HMM-based Persian continuous speech recognition system. Then pointing out the computational cost of search-based method, we proposed a linear regression process for estimating warping factor based on the scores generated by our gender detection system. Experimental results over a Persian conversational speech database shown an improvement about 0.54 percent in word recognition accuracy as well as a significant reduction in computational cost of estimating warping factor, compared to search-based approach.

Today, there are many documents on Internet, such that users can generate new documents by coping them and existing Plagiarism Detection systems (PDS) couldn't detect all kind of plagiarism. The main challenge is finding a suitable algorithm to improving the amount of similar documents and their assessing time. It’s difficult to do assessing similarity in Persian texts that different characteristics affect on it and also many of them are ambiguous. For this reason Dempster - Shefer (Evidence) theory has been used in this paper. The proposed system will assess in a two-level and in the first stage, sentences will divide in general and expert terms and then assessing by suitable measures and domain ontology. These results will be delivered to first level as "basic belief" and will be integrated by using a Dempster combination rule to create one of the second level inputs. In second level, the previous level result and another similarity measures will be weighted and combined; belief and plausibility functions for final assessment will be distinguished. This system has been used for real data assessment and compared the actual results shows that the precision between the system results and actual results is about 90%, which implies that the system can be used as Plagiarism Detection System.

In Persian language, words can be in different correct grammatical structures in a specific sentence. This makes it impossible to define a set of specific rule set for coverage all correct sentences in Persian. Consequently automatic keywords extraction is a very challenging and complicated task in Persian. In this paper, it has been tried to extract more meaningful keywords by using linguistic information and thesaurus. Using thesaurus that has a structured organization, it can be possible to improve keywords network including Equivalence Relationship, Hierarchical Relationship, and Associative Relationship. By achieving this issue, we can improve the search results of internet users and text keywords and so search generality is enhanced.In the first step, unimportant and stop words are eliminated; then the stemming is accomplished. In the following to estimate a relative importance for each word, we assign a numerical weight for each word based on a weighting strategy. The assigned value of a word indicates how important the word is in the text. Considering all the mentioned issues, specially employing a thesaurus, makes information retrieval and text classification tasks more precise. Experimental results on a multi-topic dataset indicate the efficacy of the proposed method and superiority clustering performance of our method to the ones of others.

Application of artificial neural networks (ANN) in areas such as classification of images and audio signals shows the ability of this artificial intelligence technique for solving practical problems. Construction and training of ANNs is usually a time-consuming and hard process. A suitable neural model must be able to learn the training data and also have the generalization ability. In this paper, multiple parallel populations are used for construction of ANN and evolution strategy for its training, so that in each population a particular ANN architecture is evolved. By using a bi-criteria selection method based on error and complexity of ANNs, the proposed algorithm can produce simple ANNs that have high generalization ability. To assess the performance of the algorithm, 7 benchmark classification problems have been used. It has then been compared against the existing evolutionary algorithms that train and/or construct ANNs. Experimental results show the efficiency and robustness of the proposed algorithm compared to the other methods. In this paper, the impact of parallel populations, the bi-criteria selection method, and the crossover operator on the algorithm performance has been analyzed. A key advantage of the proposed algorithm is the use of parallel computing by means of multiple populations.

The capability of the matter identification is developed considerably in hyper spectral images. The spectral reflectance of surfaces in these imaging systems in the visible and near infrared range of the electromagnetic spectrum is recorded in extremely narrow and continuous bands. But for some reasons, such as existence the mixed pixels and low spatial resolution of these images, is difficult to land cover accurate position identify. The soft classification methods provide the estimation of the membership value of various classes within mixed pixels. But, by using these methods, the matter information extraction is possible only and position information extraction in sub-pixel level is impossible. In recent years, in order to solve this problem, some methods that are called SRM, have been developed for positioning the extracted membership values by soft classification process in sub-pixels for producing a higher spatial resolution land use map. In this paper, pixel-swapping method is used as the latest SRM algorithms, and with repetition the binary case of this algorithm for each class, this algorithm has been generalized and developed for multi-class. Another main point in sub-pixel classification is the performance evaluation of these classifiers. Because of the influence of various parameters in the sub-pixel classification, the evaluation of this process is very complex. Hence, as a main and innovative activity in this paper, the Influence of the neighborhood level and the zoom factor as two important parameters in the extension pixel-swapping method has been simulated and analyzed. For this purpose, in this paper a framework for evaluating the sub-pixel classification performance based on dependent on and independent on soft classification error is proposed.

In this paper, a new unsupervised algorithm for automatic retinal blood vessel extraction is presented. A pre-processing step is introduced to eliminating optic disk and back ground noise. Blood vessel highlighting is prepared by a new method inspired by simple cells in human visual system. An adaptive threshold is introduced as an activation function of simple cells. Post-processing step is used as a final stage at the output of simple cells to eliminating exudates which is detected as the blood vessels. The results on DRIVE database demonstrate that the performance of the proposed algorithm is comparable with state-of-the-art techniques in terms of execution time and extracted vessels.