Embedded Files
Gender recognition using voice and face
Recognizing human gender plays an important role in many human computer interaction areas. For example, search engines need an image filter to determine the gender of people in images from the Internet; demographic research can use gender information extracted from images to count the number of men and women entering a shopping mall or movie theater; a “smart building” might use gender for surveillance and control of access to certain areas. Besides these kinds of broad applications, a successful and reliable gender classification approach can boost the performance of many other applications including face recognition and smart human-computer interfaces.
A system for voice and facial gender recognition has been designed, and in particular:
1) Voice: the system is able to record the voice from the microphone and make the decision in real time. The system has been tested with three databases:
a. CMU ARTIC;
b. SAVEE from Surrey University;
c. Berlin Database of Emotional Speech.
The overall accuracy of our system for voice recognition achieves a recognition rate of 98% combining all three databases together.
2) Face: two different methods have been implemented and tested using the Stanford Medical Student Face Database. The database was actually modified, cropping the faces and normalizing them in such a way that the eyes of each individual are always at the center of the image, and all the images were reduced in size (24x24) in order to prove the accuracy of our methods under low resolution data.
In the following a short description of the two methods is given:
a. The first method makes use of genetic algorithm and reduces the correlated components using the Principal Component Analysis (PCA). The 50% of the database was used for training and the remaining half was used for testing achiving a recognition rate of 92.5%;
b. In the second method, the Haar-like features are used and the database is trained and tested by the Adaboost method using 50% of the images contained into the database for each one of the two processes (training and testing). Using this method, it was achieved a recognition rate of 95%.
The entire system was developed in Matlab and a demo code (protected P-file) can be downloaded here for performance evaluation. Using the previous link, it is also possible to download the modified version of the database used for the evaluation. If you are interested in the complete version of the code, please contact me. For this project, I would like to thank Dr. Mona Fahmy for her help.
References:
[1] C.R Vimal Chand, "Face and gender Recognition Using Genetic Algorithm and Hopfield Neural Network", Global Journal of Computer Science and Technology, vol.10, April 2010.
[2] H. Lu and H. Lin, ''Gender Recognition using Adaboosted Feature", ICNC'07 (Proceedings of the Third International Conference on Natural Computation), pp. 646-65, vol. 2, 2007.
[3] B. C. Shen, C. S. Chen, H. H. Hsu, ”Fast gender recognition by using a shared-integral-image-approach,” IEEE International Conference on Acoustics, Speech and Signal Processing, pp. 521–524, April, 2009.
DVB-T2: Frame structure, coding and modulation implementation
DVB-T2 is an abbreviation for Digital Video Broadcasting – Second Generation Terrestrial and it is the extension of the television standard DVB-T.
This system transmits compressed digital audio, video, and other data in "physical layer pipes" (PLPs), using OFDM modulation with concatenated channel coding and interleaving. The higher offered bit rate, with respect to its predecessor, makes it a suited system for carrying HDTV signals on the terrestrial TV channel.
Here below, it is shown the complete block diagram for the transmitter for a DVB-T2 system:
A complete implementation of the standard ETSI EN 302 755 V1.3.1 was written in Labview programming code.
References:
[1] “ETSI EN 302 755 V1.3.1 ”, ETSI.
[2] http://en.wikipedia.org/wiki/DVB-T2
ATSC-M/H: Frame structure, coding and modulation implementation
ATSC-M/H (Advanced Television Systems Committee - Mobile/Handheld) is a standard in the USA for mobile digital TV, that allows TV broadcasts to be received by mobile devices.
Just as the DVB-H and 1seg are mobile TV extensions to the DVB-T and ISDB-T terrestrial digital TV standards respectively, ATSC-M/H is an extension to the available digital TV broadcasting standard ATSC A/53. ATSC is optimized for a fixed reception in the typical North American environment and uses 8VSB modulation. The ATSC transmission scheme is not robust enough against doppler shift and multipath radio interference in mobile environments, and is designed for highly directional fixed antennas. To overcome these issues, additional channel coding mechanisms are introduced in ATSC-M/H to protect the signal.
Here below, it is shown the complete block diagram for the transmitter for a ATSC-M/H system:
A complete implementation of the standard ATSC-M/H was written in Labview programming code.
References:
[2] http://en.wikipedia.org/wiki/ATSC-M/H
Genetic algorithm to repair a 2x1 homogeneous network iterated N times
A homogeneous network is defined as a network composed of similar nodes. The simpler example of an homogeneous network is composed of 2 nodes as below:
A network is said to be completely iterated if each element in the network is replaced by a copy of the original network, while maintaining an identical interconnection structure.
Using genetic algorithm, it is created an application that given the stuck-fault in a 2x1 homogeneous network iterated N times, it is able to repear it in order to have the original behaviour, when it is possible, changing the output function of the nodes, according to the following table:
The executable of the implementation of the application written in Labview programming code can be found here.
References:
[1] A. Noore and R.S.Nutter, “On Testing Iterated Neural Network Structures,” IEEE Southeastern Symposium on System Theory, Columbia, SC, March 10-12, 1991.
[2] R.H. Urbano, “On the Convergence and Ultimate Reliability of Iterated Neural Nets”, IEEE Transactions on Electronic Computers, EC-13 (3), pg. 204-225, June 1964.
Google Sites
Report abuse