THE CSI JOURNAL ON COMPUTER SCIENCE AND ENGINEERING
Year:2006 | Volume:4 | Issue:1 (A)|Papers Count:6

This paper presents an expert system for emotion recognition from face image sequences. Such a system can be used for human computer interaction. Input to the system is frontal face image sequences, in which it is assumed that no rigid head motion is present and intensity is uniformly distributed. The system tracks a set of manually marked points of the face and classifies facial feature variations in terms of Action Units. Later, the system classifies detected action units in terms of a subset of basic emotions, i.e. joy, surprise, sadness, inattentiveness, anger, disgust and fear. For each action unit and emotion category, the intensity criteria is defined and used for merging similar action units and emotions. The system recognizes a combination of emotion categories and resolves possible contradictions (e.g. sadness and joy) by eliminating lower intensity emotion. Overall emotion recognition accuracy is about 80%.

In this paper a question answering system based on ontology has been introduced. In fact, we consider this system with respect to the important of information retrieval from the huge amount of information resources and capability of the natural language processing techniques. Also, because of the lack of the semantic processing of the most existing question answering systems, in this paper, the ontology has been applied as an approach for concept identification and discovering conceptual relationship between available concepts in a domain. The efficiency of the proposed question answering system was evaluated by applying the proposed system to a document retrieval problem and we saw that using ontology, the accuracy of the answering increased. Also, we observed that, with increasing the number of the ontology concept nodes, the accuracy of the question answering was increased.

Handwritten text recognition is one of the most interesting and open problems in optical character recognition. A variety of writing patterns, styles, and pen tips make this problem even more complicated. Using fuzzy logic techniques is shown to improve the performance of recognizing algorithm. In this research, a new fuzzy inference system for recognition of Persian-Arabic handwritten digits has been proposed. It is based on "line sweeping" of segmented image, which we called it hit counting. The system extracts a digit from the input image, and finally recognizes the digit. The proposed system does not need any preprocessing algorithm like skeletonizing or contouring. It is also independent of thickness and size of pen while it is robust to small rotations. The implemented system could achieve 94.8% recognition on Persian-Arabic handwritten digits dataset as well as complete recognition for specific typefaces.

The role of designers is very important in relational database schema design. Relational database design tools help designers to design semantically correct and without anomaly databases; however, they are not good enough yet. There is a shortcoming in quality evaluation techniques for relational databases; therefore, database designers may not be able to systematically design better database schemes. In this paper, first, we define quality indicators for evaluating quality of relational database schemes based on software measurement techniques. Then we introduce a technique for evolutionary design of relational database schemes based on the model with which the quality of database schemes can be estimated. In this technique we try to minimize the number of states to be considered for finding the suitable answer.

Compressors are widely used in realization of multi-operand addition as well as the reduction of multiplication tree. One of the major properties of 4-2 compressor is compression degree of two. 4-2 compressors are used in order to design and implement larger compressors such as 6-2, 9-2 and IEEE 754 standard (27-2 compressor). The inconvenient of conventional 4-2 compressor is the incapability of handling the signed numbers. In this paper a new technique for addition is proposed. Using this method and current mode multiple valued logic we propose a new circuit. It enjoys the 2's complement property and preserves the conventional 4-2 compressor reduction degree. This design is simpler to realize. The propagation delay in the best and worst cases is exactly the same. The number of transistor used is reduced which results in a reduction of power dissipation and chip area. The related circuits are simulated with HSPICE using 0.25mm technology. The simulation results confirm what we have darned.