Ecole supérieu​re d'informaitque Salama

Features handling in computer vision: Theory, applications and Challenges

Presenter: Prof. Jules R Tapamo

School of Engineering
University of KwaZulu-Natal
Durban, South Africa


Computer vison aims at mimicking the human vision ability. This process starts by using sensors to capture raw digital images or videos from which relevant information is extracted to gain high-level understanding of the world around us. Computer vision has become an essential field in modern artificial intelligence and machine learning applications. Features extraction and features learning are two fundamental methods used for extracting relevant information from images or videos Features extraction involves manually designing algorithms to extract specific features such as edges, textures, or colors, from images or videos. In contrast, feature learning is a machine learning approach that allows algorithms to automatically learn features from raw data without the need for manual feature engineering. While feature extraction provides better interpretability and control over the features extracted, feature learning has produced in many instances’ better performance and generalization on new data. However, feature learning requires large amounts of annotated data and significant computational resources. we will explore the fundamental concepts of features extraction and features learning in computer vision. We will further discuss the theories behind both methods and the advantages and disadvantages of each approach. Applications of these techniques are diverse and include object recognition, image segmentation, facial recognition, and medical imaging. However, several challenges remain in both methods, such as overfitting, generalization, and the trade-off between computational resources and performance. Additionally, we will focus on the recent advances in deep learning techniques, specifically convolutional neural networks, and their successful application to feature learning in computer vision. We will highlight the improvements in performance and efficiency of these methods in various applications. It will also be shown how in recent years, deep learning techniques, such as convolutional neural networks, have been successfully applied to feature learning in computer vision, leading to significant improvements in performance in various applications. We will also highlight some further research that are required to address the remaining challenges and improve the performance and efficiency of both feature extraction and feature learning methods in computer vision.


Jules R Tapamo is Professor of Computer Science and Engineering in the School of Engineering at The University of KwaZulu-Natal. Dr. Tapamo received his PhD in Computer Science and his DEA in Applied Mathematics and Computer Science, both from the University of Rouen, France. He is Academic Leader: Teaching and Learning in the School of Engineering at UKZN. He is Invited Professor at the Sudan University of Science and Technology. Dr Tapamo's main research interests lie in the fields of image processing, Computer vision, Machine Learning, Biometrics and Data Science. His key focuses are on behavior detection and modelling, feature extraction and learning from surface and scene and extraction of knowledge from unstructured data. He is involved in cross-disciplinary research on medical informatics, safety and security, remote sensing, agriculture, pollution, power distribution infrastructure monitoring. He is IEEE, IEEE Computer Society, IEEE Signal Processing Society, IEEE Computational Intelligence society, IEEE Geoscience and Remote Sensing Society. Dr Tapamo is rated Scientist of the South African National Research Foundation. He has been involved in several local and international conference committees. He is also reviewer of several international and local journal.