M.Sc. Engg.

Browse

Now showing 1 - 3 of 3

Improvement of Face Detection Incorporating Illumination-based Robust Skin Color Measure
(Khulna University of Engineering & Technology (KUET), Khulna, Bangladesh, 2019-06) Akash, Md. Asif Anjum; Akhand, Prof. Dr. Muhammad Aminul Haque
Human vision system is amazing in detecting face easily but it is very challenging in computer vision and image processing as it depends on quality of image, illumination, lighting conditions, face sizes, occlusions, and face position etc. The existing face detection systems, including popular Haar feature based face detection (HFFD), very often detect a region as a face which is eventually not a face. To counteract such false detection, incorporation of human skin color property is considered a way of improving face detection accuracy in several recent studies. But these methods are found to be dependent on illumination conditions meaning that the performances of these methods degrade when applied to images with different illumination conditions. The aim of this study is to devise a robust face detection system integrating skin color matching that will perform well under different illumination conditions. In pursuit of this goal, a novel skin color matching method is proposed which is a composite of two rules to balance the high and low intensity facial images by individual rule. In the proposed method, illumination intensity of a given facial area is measured and then appropriate rule is applied based intensity value to verify the area as face or not. The proposed skin color matching is verified in face detection with HFFD on four benchmark face datasets (Put, Caltech, Bao and Muct) and a self-prepared dataset. Experimental results and analysis revealed the effectiveness of proposed composite skin color matching to improve face detection while compared with prominent existing skin color-based face detection methods.
Speech Enhancement Using Convolutional Denoising Auto-Encoder
(Khulna University of Engineering & Technology (KUET), Khulna, Bangladesh, 2019-05) Shahriyar, Shaikh Akib; Akhand, Prof. Dr. Muhammad Aminul Haque
Speech signals are complex in nature with respect to other forms of communication media such as text or image. Different forms of noises (e.g., additive noise, channel noise, babble noise) interfere with the speech signals and drastically hamper the quality of the speech. Enhancement of speech signals is a daunting task considering multiple forms of noises while denoising a speech signal. Certain analog noise eliminator models have been studied over the years for this purpose. Researchers have also delved into some machine learning techniques (e.g., artificial neural network) to enhance speech signals. In this study, a speech enhancement system is investigated using Convolutional Denoising Autoencoder (CDAE). Convolutional neural network (CNN) is a special kind of deep neural networks which is suitable for 2D structured input (e.g., image) and CDAE is a CNN based special kind of Denoising Autoencoder. CDAE takes advantages from the 2D structured inputs of the features extracted from speech signals and also considers the local temporal relationship among the features. In the proposed system, CDAE is trained considering features from noisy speech signal as input and clean speech features as desired output. The proposed CDAE based method has been tested on a benchmark dataset, called Speech Command Dataset, and attained 80% similarity between denoised speech and actual clean speech. The proposed system achieved perceptual evaluation of speech quality (PESQ) value of 2.43 which outperformed other related existing methods.
University Course Scheduling using Prominent Nature Inspired Techniques
(Khulna University of Engineering & Technology (KUET), Khulna, Bangladesh, 2019-04) Hossain, Sk. Imran; Akhand, Prof. Dr. Muhammad Aminul Haque
The University Course Scheduling Problem (UCSP) is a highly constrained real-world combinatorial optimization task. Solving UCSP means creating an optimal course schedule by assigning courses to specific rooms, instructors, students, and timeslots by taking into account the given constraints. Several studies have reported different metaheuristic approaches for solving UCSP including Genetic Algorithm (GA) and Harmony Search (HS) algorithm. Various Swarm Intelligence (SI) optimization methods have also been investigated for UCSP in recent times and a few Particle Swarm Optimization (PSO) based methods among them with different adaptations are shown to be effective. In this study, two novel PSO and Group Search Optimizer (GSO) based methods are investigated for solving highly constrained UCSP in which basic PSO and GSO operations are transformed to tackle combinatorial optimization task of UCSP and a few new operations are introduced to PSO and GSO to solve UCSP efficiently. In the proposed methods, swap sequence-based velocity and movement computation and its application are developed to transform individual particles and members in order to improve them. Selective search and forceful swap operation with repair mechanism are the additional new operations in the proposed methods for updating particles and members with calculated swap sequences. The proposed PSO with selective search (PSOSS) and GSO with selective search (GSOSS) methods have been tested on an instance of UCSP resembling the course structure of the Computer Science and Engineering Department of Khulna University of Engineering & Technology which has many hard and soft constraints. Experimental results revealed the effectiveness and the superiority of the proposed methods compared to other prominent metaheuristic methods (e.g., GA, HS).