Document Type : Original Article
Assistant Prof., Faculty of Electronic Engineering, Comm. Eng. Dept., Menouf. Egypt.
Assistant Lecturer., Faculty of Electronic Engineering, Comm. Eng. Dept., Menouf. Egypt.
Associate Prof., Faculty of Electronic Engineering, Comm. Eng. Dept., Menouf. Egypt.
Prof., Faculty of Electronic Engineering, Comm. Eng. Dept., Menouf. Egypt.
Image data compression is used to reduce the transmission rates or the amount of informatipn to be sent or stored without greatly affecting the quality of the reconstructed images. Many techniques, such as vector quantization, have been used in'order to satisfy these requirements. Using the neural network techniques, rather than the traditional technique in this subject decreases the loss of image information and hence enhances its quality specially at low bit rates.
Some of the neural network models proposed for image data compression still have some defects specially, when the actual images are not included in the training phase of the network. In this paper a new training set is proposed to be used in the training set of the Kohonen self-organizing map when it used in image data compression applications to increase its efficiency. This proposed training set is statistically dependent with all images independent of their types. A predictive vector quantization using both Kohonen self-organizing map and the adaptive differential pulse code modulation with the linear neural network predictor is also introduced. The simulation results':this paper show that the performance of the proposed techniques is much better than that used by others.