Haar Walsh, Slant, and Discrete Cosine transform coding of images  A comparison
Abstract
The behavior of frequently used transform matrices for the transformation of image data is compared. An analog video signal is sampled and converted to digital data. Each picture can then be regarded as a matrix with integer entries. All matrices usually employed in transform coding systems are orthonormal matrices. An orthonormal transformation is energyinvariant and errorinvariant. Attention is given to aspects of block quantization and coding, inverse transformation, rounding, digital to analog conversion, the definition and properties of several orthonormal transforms, the degree of complexity of a hardware realization of the arithmetic unit, and the normalized mean square error. The order obtained with respect to increasing quality is Harr, Walsh, Slant, Discrete Cosine, and KarhunenLoeve transform.
 Publication:

Electromagnetic Compatibility
 Pub Date:
 1977
 Bibcode:
 1977elco.proc...37M
 Keywords:

 Cosine Series;
 Data Sampling;
 Image Processing;
 Signal Encoding;
 Transformations (Mathematics);
 Walsh Function;
 Computerized Simulation;
 Digital To Analog Converters;
 Matrices (Mathematics);
 Network Synthesis;
 Orthogonal Functions;
 RootMeanSquare Errors;
 Sine Waves;
 Video Data;
 Waveforms;
 Communications and Radar