An optimized approach for efficient transmission of an artificial neural network processed images over the internet

Abstract

Internet growth in recent years has encouraged many new applications to be provided for the user such as video conferencing and video telephony. Most of these new applications require data involving images characterized by a large number of bits. Therefore, it is a difficult task to implement these types of applications since the data needs to be transferred from one place to another over the network. Moreover, these applications are intended to be real time applications implying that they are very sensitive to delay and jitter which add to the challenges of providing such applications. Consequently, two strategies are needed in order to meet the challenges of transferring and storing very high volume of data for the application while maintaining strict requirement for delay and jitter. The first strategy involves qualitative and quantitative optimization of the application’s data. The qualitative optimization ensures that the data of the application meets at least a minimum quality requirement (e.g. image resolution). Quantitative optimization is accomplished by using compression techniques to reduce the number of bits that are required to be transmitted over the network. The second strategy involves optimizing the utilization and performance of the network. This research, which consists of two phases, examines an optimized approach for transferring and storing images. Phase one deals with the processing and compressing of the images. Processing the images is done by applying an adaptive filter in order to enhance the visual quality while compression is used to reduce the amount of data needed to be stored or transmitted. The study developed a Multilayer Feed Forward Artificial Neural Network (MFFANN) for image compression. Based on Gradient Descent, Conjugate Gradient, Quasi-Newton techniques, three different error back propagation algorithms were developed for use in training the MFFANN. The essence of this study is to investigate the most efficient and effective training methods for use in image compression and its subsequent applications. The results show that the Quasi-Newton based algorithm has a better performance as compared to the other two algorithms. In Phase two, the study proposes a new admission control mechanism that aims to enable Multi Protocol Label Switching (MPLS) tunnel differentiations. It also presents a simulation based evaluation for the proposed mechanism. The results demonstrate the effectiveness of the proposal, as it is able to maintain a robust and stable end-to-end quality of service for selected flows, which leads to a better performance in terms of throughput, delay and jitter. The implementation of the optimization approach presented in this research could be highly beneficial in providing online real time services such as medical services, where consultation or even remote medical operations could take place using the Internet.