Edge Preserving Image Compression Technique using Adaptive Feed Forward Neural Network

The aim of the paper is to develop an edge preserving image compression technique using one hidden layer feed forward neural network of which the neurons are determined adaptively. Edge detection and multi-level thresholding operations are applied to reduce the image size significantly. The processed image block is fed as single input pattern while single output pattern has been constructed from the original image unlike other neural network based techniques where multiple image blocks are fed to train the network. The paper proposes initialization of weights between the input and lone hidden layer by transforming pixel coordinates of the input pattern block into its equivalent one-dimensional representation. Initialization process exhibits better rate of convergence of the back propagation training algorithm compare to the randomization of initial weights. The proposed scheme has been demonstrated through several experiments including Lena that show very promising results in compression as well as in reconstructed images over conventional neural network based techniques available in the literature.