Title: A GPU Based Efficient Trademark Retrieval Technique using a Weighted Combination of Multiple Image Features, IEEE Conference on Communication, Science & Information Engineering (CCSIE), London, UK, 25-27 July, 2011.
Authors:Madeena Sultana, Nurul Muntasir Mamun, Mohammad Shorif Uddin, and Maaruf Ali
Abstract:Accuracy and computation efficiency are two main criteria for retrieval of an object from large image databases. Recently, we proposed a technique using weighted combination of multiple image features, such as color histogram, distance, and moment features which has shown its effectiveness and robustness in retrieval of trademark images from large databases. However, a limitation of this algorithm is high computational burden on processing large image datasets. Current graphics processing units (GPUs) show tremendous computation power with parallel processing. In this paper, we present a GPU based implementation of our technique on NVIDIA’s Compute Unified Device Architecture (CUDA) platform. We have considered a large database containing 1500 trademark images and achieved more than 50% (2.5x speedup) reduction in computation time over usual CPU time.
File:CCSIE_Madeena_paper.pdfLink:Keywords:Image retrieval; color histogram; moment invariant; trademark image database; CUDA; GPU
Title:Computational Speed-up in Digital Holographic Image Reconstruction using Graphics Processing Unit, 17th Mathematics Conference, Jahangirnagar University, Dhaka, Bangladesh, 22-24 December, 2011.
Authors:Mohammad Shorif Uddin and Madeena Sultana
Abstract:File:Link:Keywords:
Title:Simple and Fast Implementation of SegmentedMatrix Algorithm for Haar DWT on a Low Cost GPU, Proc. of Int. Conf. on Advances in Computer Science 2011, pp. 103-106, 20-21, Noida, India, 2011.
Authors:Madeena Sultana and Nurul Muntasir Mamun
Abstract:Haar discrete wavelet transform (DWT), the simplest among all DWTs, has diverse applications in signal and image processing fields. A traditional approach for 2D Haar DWT is 1D row operation followed by and 1D column operation. In 2002, Chen and Liao presented a fast algorithm for 2D Haar DWT based on segmented matrix. However, this method is infeasible for its high computational requirements for processing large sized images. In this paper, we have implemented the segmented matrix algorithm on a low cost NVIDIA’s GPU to achieve speedup in computation. The efficiency of our GPU based implementation is measured and
compared with CPU based algorithms. Our experimental results show performance improvement over a factor of 28.5 compared with Chen and Liao’s CPU based segmented matrix algorithm and a factor of 8 compared to MATLAB’s wavelet function for an image of size 2560×2560.
File:Link:http://doi.searchdl.org/02.ACS.2011.02.117Keywords:Haar discrete wavelet transform (DWT), CUDA, GPU, segmented matrix algorithm, parallel discrete wavelet transform
Title:Fast Holographic Image Reconstruction using Graphics Processing Unit, ULAB Journal of Science and Engineering, vol. 2, no. 1, pp. 35-41, November 2011.
Authors:Mohammad Shorif Uddin, Madeena Sultana, and Md. Ziarul Islam
Abstract:File:Link:Keywords:
Title:Aiding Autistic Children Employing Computer Technology, International Conference on Medical Physics in Radiation Oncology and Imaging (ICMPROI-2011), 11-13 March, 2011, Gono Bishwabidyalay (University), Dhaka .
Authors:Hanif Ali, Mohammad Shorif Uddin and Madeena Sultana
Abstract:
File:Link:Keywords:
Title:Trademark Recognition using a Weighted Combination of Different Image Features, Proc. of 3rd IEEE International Conference on Machine Vision (ICMV), ISBN, 978-1-4244-8888-9, pp. 357-360, Hong Kong, 2010.
Authors:Madeena Sultana and Mohammad Shorif Uddin
Abstract:File:Link:http://www.ijcte.org/icmv/icmv2010.htmKeywords:
Title:Impact of Interferers on Performance of Linear Highway Mobile Cellular Network, Proceeding of Bangladesh Electronics Society, ISBN 984-300-000645-7, 2007
Authors:Md. Imdadul Islam, Jamilur Rahman, Madeena Sultana
Abstract:For high way mobile cellular network cells are arranged along the high way, therefore number of interferer cells for a call is two instead of six cells of planar network. In channel borrowing scheme channels are borrowed from adjacent under loaded cells to support an overloaded cell. In dynamic channel allocation scheme channels are distributed centrally according to traffic demand of individual cell. In both scheme presence of one user creates interferences to other users and some times some calls have to be terminated to keep the signal to interference ratio below the threshold level. Traffic analysis of such network is done based on two dimensional Markovian chain where on offered traffic is related to arrival rate of successful users and the other is related to that of interferer users. In mobile cellular network combined new and handover call arrival follow Erlang’s probability distribution but number of interferer users around a cell is limited therefore traffic of interfere users follow limited user traffic i.e. Engset traffic model. In this paper a two dimensional Markovian chain is designed taking a number of 50 interferer users and 7 number of channels and the Markovian chain is solved in tabular form to find the Blocking Probablity of interfering users.
File:Link:Keywords:
| Webmail | OPAC | Site Map