Codebook enhancement in vector quantization image compression using backpropagation neural network

Vector Quantization (VQ) is a powerful technique for image compression. One of the most VQ problems is the high computational complexity of searching for the closest codevector in the codebook during the compression phase. Artificial Neural Networks (ANNs) were used in image compression where high computational performance is required. A three layered Back Propagation Neural Network (BPNN) was proposed in this research for building an enhanced codebook for VQ compression of images. The Backpropagation neural network algorithm (BP) was used for training the designed VQ BPNN. Finally, a trained VQ BPNN was obtained to produce the codevectores of any image by the hidden layer neurons. We can later apply both trained and un-trained images to this VQ BPNN to compress them. Experiments were conducted with different VQ BPNN architecture and BP parameters to speed up this algorithm and enhance VQ codebook. It is observed that proposed algorithm is faster than other algorithms, although it needs a time for learning process. The performance of VQ BPNN image compression can be increased by modifying the VQ BPNN architecture especially, the number of hidden layer neurons and modifying BP learning parameters. © 2011 Asian Network for Scientific Information.