Gouda I. Salama

Work place: Military technical college; /Department of computer engineering, Cairo, Egypt

E-mail: gisalama@mtc.edu.eg


Research Interests: Image Processing, Image Manipulation, Image Compression, Computer Graphics and Visualization, Pattern Recognition


Gouda I. Salama, received the Bachelor engineering and Masters' engineering degrees from MTC, Cairo, Egypt, in 1988 and 1994, respectively. As well, he received the Ph.D. degree in Electrical and computer engineering from Virginia Tech. University, U.S.A., in 1999. He is currently a faculty member with the Department of Computer Engineering, MTC. His research interests are in image and video processing, pattern recognition, and information security.

Author Articles
Arabic Opinion Mining Using Combined CNN - LSTM Models

By Hossam Elzayady Khaled M. Badran Gouda I. Salama

DOI: https://doi.org/10.5815/ijisa.2020.04.03, Pub. Date: 8 Aug. 2020

In the last few years, Sentiment Analysis regarding customers' reviews in order to comprehend the opinion polarity on social media has received considerable attention. However, the improvement of deep learning for sentiment analysis relating to customer reviews in Arabic language has received less attention. In fact, many users post and jot down their reviews in Arabic daily, so we ought to shed more light on Arabic sentiment analysis. Most likely all previous work depends on conventional classification techniques, such as KNN, Naïve Bayes (NB), etc. But in this work, we implement two deep learning models: Long Short Term Memory (LSTM) and Convolution Neural Networks (CNN), in addition to three traditional techniques: Naïve Bayes, K-Nearest Neighbor (KNN), Decision trees for sentiment analysis and compared the experimental results. Also, we offer a combined model from CNN and Recurrent Neural Network (RNN) architecture where this model collects local features through CNN as the input for RNN for Arabic sentiment analysis of short texts. An appropriate data preparation has been conducted for each utilized dataset. Our Conducted experiments for each dataset against traditional machine learning classifier; KNN, NB, and decision trees and regular deep learning models; CNN and LSTM, has resulted in impressive performance using our proposed combined (CNN-LSTM) model with an average accuracy of 85,83%, 86,88% for HTL and LABR datasets respectively.

[...] Read more.
An Adaptive Deblocking Filter to Improve the Quality of the HEVC Standard

By Alaa F. Eldeken Gouda I. Salama

DOI: https://doi.org/10.5815/ijigsp.2015.03.02, Pub. Date: 8 Feb. 2015

In this paper, we present an adaptive deblocking filter to improve the video quality for high efficiency video coding (HEVC) scheme. The HEVC standard is a hybrid coding scheme using block-based prediction and transform encoding/decoding. At the decoding step, the boundary of any two adjacent blocks causes visual discontinuities called blocking artifacts that can be removed using deblocking filter. Conventional approaches, including the HEVC standard, tend to remove those artifacts using two offset parameters that are defaulted to 0. However, such a choice is not necessarily suitable to encode/decode all video sequences. The proposed approach reduces an exhaustive search among a set of candidate offsets to eventually select the best offsets adaptively (i.e., for each frame) according to some characteristics of the data sequences. Improvements are shown using the proposed approach in terms of rate-distortion (RD) performance as opposed to the HEVC standard without changing the compression ratio and with negligible change in the encoding/decoding time.

[...] Read more.
Other Articles