BRAC University Institutional Repository
Preserving Knowledge, Advancing Research, Sharing Scholarship
A digital platform for collecting, preserving, and sharing BRAC University’s scholarly, academic, and institutional outputs.
Contribute to the Repository
Faculty members and students are invited to submit their research publications, theses, dissertations, and scholarly works to increase visibility, access, and long-term preservation.
Recent Submissions
Enhancing facial image resolution: Leveraging UNET++ for super-resolution using deep learning
(Institute of Electrical and Electronics Engineers Inc., 2023-01-01) Ajwad, Asef Jamil; Rafid, Sk Tahmed Salim; Podder, Saurov; Department of Computer Science and Engineering
This paper presents a novel and robust method for achieving 4X upscaling of facial images, leveraging the UNET++ architecture and applied to the widely recognized Flickr-Faces-HQ (FFHQ) dataset comprising high-resolution 512x512 images. The study systematically explores the performance of four distinct UNET++ model sizes, each offering a unique balance between computational efficiency and upscaling quality. Through extensive experimentation, the proposed approach demonstrates remarkable prowess in preserving intricate facial features, enhancing texture details, and maintaining visual realism during the upscaling process. Notably, the best-performing model exhibits an impressive average Peak Signal-to-Noise Ratio (PSNR) of 30.465 dB, a Structural Similarity Index (SSIM) of 0.859 and a Multiscale SSIM (MS-SSIM) of 0.976, establishing a new standard of excellence in facial image upscaling. These findings contribute to the advancement of image processing techniques, particularly in the context of facial imagery, and hold profound implications for a wide range of applications, including computer graphics, image enhancement, and face recognition systems.
ReSkipNet: Skip connected convolutional autoencoder for original document denoising
(Institute of Electrical and Electronics Engineers Inc., 2023-01-01) Rahman, Mohammad Muhibur; Ahmed, Anushua; Hasan Mahin, Mohammad Rakibul; Bin Kibria, Fahmid; Moonwar, Waheed; Rhythm, Ehsanur Rahman; Alim Rasel, Annajiat; Department of Computer Science and Engineering
Data pre-processing, data analysis, and Optical Character Recognition need a huge amount of clean data, and document images are usually a good source for this. However, document images frequently exhibit blurring and various other forms of noise, which can pose challenges in their manipulation and analysis. To denoise and deblur such document images, autoencoders have been used for a long time. For this task, we propose a novel Convolutional Autoencoder Network which is composed of multiple skip-connected residual blocks and other layers for supporting the encoder and decoder parts. This model not only uses less computational power to denoise existing document image datasets but also performs well. While prior research primarily concentrates on optimizing evaluation metrics, our approach additionally prioritizes larger resolution input sizes. This characteristic of using larger image sizes enhances its practicality and usability as real-world documents are typically characterized by a higher word density. Moreover, in order to further advance the development of our model, we produced an original dataset and proceeded to train our model on this dataset, resulting in satisfactory outcomes.
Towards safer roads: Raspberry pi-driven drowsy driver detection using EAR analysis
(Institute of Electrical and Electronics Engineers Inc., 2023-01-01) Zillur, Sheikh Faiyadh; Zaman, Srejoni ; Tasfia, Faiza Zahin ; Rafid, Sk Tahmed Salim; Sumaya, Atoshe Islam; Huda, A. S. Nazmul ; Department of Computer Science and Engineering
The detection of drowsiness is a safety measure that can prevent road accidents caused by drivers who fell asleep behind the wheel. This article develops a driver drowsiness detection system which is based on Raspberry Pi processor and "facial-68-landmark"model for facial feature extraction. The developed system is extensively evaluated through various test cases, yielding valuable insights into its performance and accuracy. By utilizing the Eye Aspect Ratio (EAR), the system establishes drowsiness thresholds, identifying an EAR below 0.25 as indicative of drowsiness and an EAR below 0.15 as a sign of sleepiness. In order to assess the performance of the proposed system, test results for 10 volunteers is considered. The study reveals that the system performs optimally with a direct camera placement pointed towards the driver, achieving an impressive accuracy of 99% when drivers do not wear glasses. Under different conditions including spectacles and viewing angles variation, the system maintains an average accuracy of 97% across 600 tested frames. These results prove that the developed system can be applied on real world applications.
Automated image caption generation using deep learning
(Institute of Electrical and Electronics Engineers Inc., 2023-01-01) Hasan, Mahmudul; Prithila, Sara Jerin; Al Ahasan, Tamim; Hassan, Mahmudul; Hossain, Abid; Bhuiyan, Sania Azhmee; Alim Rasel, Annajiat; Department of Computer Science and Engineering
Image captioning means generating relevant texts from images that describe the image. With the growing advancement of deep learning, automatic image caption generation has become an exciting problem among researchers. Image captioning is used in various fields of computer science including computer vision and NLP. It is helpful in many cases such as visual impairment. Though many research works have already been published on this topic using different deep learning models, we can work with various features of the deep learning models to achieve better accuracy. In this paper, we are focusing on combining several deep-learning models to get the desired accuracy. VGG16 model is used which is a CNN architecture to extract essential features from the image. For the generation of relevant captions, LSTM is adopted rather than RNNs as LSTM generates better captions compared to RNNs and it consumes less time. Finally, the model is trained on Flickr 8k datasets.
Bacterial diversity with emerging antimicrobial resistance of diabetic foot ulceration and current detection techniques: A review
(Modestum LTD, 2021-01-01) Ahmed, Akash; Alvi, Sayeed Akhtar; Aftab I.B.; Akhtar, Fahmina
Diabetic foot ulcer (DFU) is becoming one of the major complications around the world with associated consequences such as lower-extremity amputation, high morbidity, mortality and hospitalization. It has the possibility to become the next global epidemic. Major part of the concern comes from the diverse bacterial and fungal population that is found at the infection site and their growing antimicrobial resistance. If the threat of antimicrobial resistance is not dealt with than it will rise to become the main cause of mortality and below knee amputation in case DFU. Also, most of the time main focus is given on detecting bacterial population which causes the fungal population to go unnoticed and act as the silent enemy. Bacterial and fungal prevalence scenario from different countries have been discussed in this study along with the alarming antibiotic resistance scenario around the globe. Furthermore, choosing the correct technique to identify them also plays a vital role. With proven lacking’s of the culture-based methods maybe it is time to move on to the faster and more specific molecular methods. As, many of the molecular techniques have already proven to be more efficient. This review discussed the bacterial and fungal prevalence along with their growing antimicrobial resistance and evaluated different biochemical and molecular techniques in identification process.