Two novel speech coders , called flag modulo pulse code modulation GMPCM and enhansened flag modulo pulse code modulation EGMPCM , are presented. The coders are designed for medium and high bit rate applications. These coders use the modulo PCM principle. A flag bit is used to indicate changing in speech segments. The coders have small modulo amplitude and hence better SNR. Dynamic quantization and bit allocation are also used in these coders in effect these coders function as dynamic quantizers. EGMPCM coders use a new technique to reduce the modulo amplitude by modifying the input speech files. Both coders do not use predictors neither in coding nor in decoding stages. The coders depend only upon the actual difference between conp.

Approximately linear phase IIR digital filters are designed here as the sum of two allpass transfer functions, one of them is an exactly linear phase FIR filter or simply a delay element, while the other one is an IIR allpass. The IIR allpass parameters are obtained by using an optimization technique, in which the stability of the transfer function is guaranteed through special choice of the optimisation parameters. The method has the advantage that it can be easily applied to switched capacitor filters. A new simple implementation method is given for this case

In practical vector quantization (VQ) of images, the used pixel block dimensions are kept small to reduce the cost of computations. This in turn results in highly correlated blocks and the corresponding VQ indices will inherit this high correlation. The compression of the basic VQ can be increased through utilising this high correlation of indices by inserting a lossless index compression stage after the VQ stage. A new index compression algorithm is introduced. In this algorithm the 2 dimensional index map is divided into nonoverlapping square blocks. Index usage in each of these blocks is employed to remap (renumber) the reduced number of actually used indices in this block, thus resulting in reduced bit rate expressed in bits/pixel The proposed algorithm reduces the average bit rate by a value depending on the codebook size, namely a reduction of about 32% for codebook size of 64, and down to about 23% for codebook size of 1024. Moreover this algorithm lends itself to being cascaded by other index compression algorithms resulting in increased compression.

In this paper, we propose an algorithm for the application of one-hidden layer Feedforward Neural Network (OHL-FNN) to image compression. The algorithm combines OHL-FNN with Discrete Cosine Transform (DCT), here, the neural network learning algorithm performs the compression in a spectrum domain of DCT coefficients, i.e., the OHL-FNN approximates only the DCT coefficients representing the high detailed part of the image, Network parameters are stored in order to recover the image. Results, compared with baseline JPEG algorithm, demonstrate that the new algorithm dramatically increase compression for a given quality; conversely it increases image quality for a given compression ratio.

Image compression using vector quantization (VQ) results in highly correlated indices. The correlation between these indices is used to reduce the bits needed to represent them. This is done by many index compression algorithms such as the Hu and Chang, search order coding (SOC), and switching tree coding (STC). A new algorithm for VQ index compression is introduced and it utilizes the local statistics of each image and the repeating pattern of its adjacent indices. The proposed algorithm improves the index compression performance of the basic VQ, with a relatively slight increase of complexity.