Summary form only given. We post-process images compressed by lossy JPEG algorithms (DCT and LOGO), following decompression, using Bezier polynomials for enhancement. Both JPEG algorithms, namely the DCT-based JPEG and the near-lossless baseline JPEG-LS (aka LOGO), employ spatial quantization. The DCT-based version introduces an artifact known as mosaic or "blockiness" by coarsely quantizing the DC coefficients. By quantizing the prediction errors, JPEG-LS introduces a "banding" artifact specially noticeable for quantization coarser than "8-off". The human visual system (HVS) is specially sensitive to degradations on smooth regions in the image. We propose a post-processing algorithm based on blending for enhancement of the degraded smooth regions due to quantization. In this work, we use Bezier polynomials of variable orders. Enhancement is obtained by blending each pixel in low-activity regions so the artifacts are reduced. The coefficients of the Bezier polynomials for the various orders used (from 3rd to 7th order) are precomputed and stored in look-up tables for speed. Active pixels, detected by using a threshold of 3/2 of the quantization step size, are left alone. The threshold is theoretically justified, avoids the problem of threshold determination on ordinary segmentation, and excludes quantization noise. In order to avoid over-smoothing of large low activity regions, the Bezier polynomials' order used to blend each pixel P/sub o,j/ is variable. Results are presented for lossy JPEG and JPEG-LS images for very low bit-rates.

This comment points out some omissions in the work of Duttweiler and Chazmas (see ibid., vol.4, p.237, 1995) on credit related to scaled-count estimators that perform count-scaling when the count of the less probable symbol (LPS) reaches a limiting value.

The compression method is based on composing phrases from symbols. The authors extend the sort-order property to parsing models, i.e. to Variable-to-Fixed Length codes, or a static Ziv-Lempel algorithm, or alternatively a Tunstall algorithm for an adjoint source. The parsed phrases comprising the original storage data units have the same position in the sort ordering as the original units themselves. The VFL result may be further compressed by use of Variable-to-Variable Length techniques based on the relative frequencies of the parsed phrases. The sort-order property is facilitated by an 'end of record' symbol and requires a new zilch symbol.>

The JPEG lossless arithmetic coding algorithm and a predecessor algorithm called Sunset both employ adaptive arithmetic coding with the context model and parameter reduction approach of Todd et al. The authors compare the Sunset and JPEG context models for the lossless compression of gray-scale images, and derive new algorithms based on the strengths of each. The context model and binarization tree variations are compared in terms of their speed (the number of binary encodings required per test image) and their compression gain. In this study, the Bostelmann (1974) technique is studied for use at all resolutions, whereas in the arithmetic coded JPEG lossless, the technique is applied only at the 16-bit per pixel resolution.>

Developments related to the Q-Coder approach to adaptive binary arithmetic coding are reported. A theorem is established and more probable symbol renorms per less probable symbol at the equilibrium point are plotted for fixed and random augends. Details of Q-Coder evolution into the more easily implemented single-adaptor version are provided.>

We describe a one-pass compression scheme which presumes no statistical properties of the data being compressed. The model structure adaptively selects a subset of first-order Markov contexts, based on an estimate of the candidate context's popularity. The probability distributions for the unselected (lumped) first-order contexts are made the same, reducing cost over a full first-order Markov model. Symbol repetitions are handled in special secondorder Markov contexts. The statistics for each symbol are adaptively determined by an extension of earlier work.

The Ziv-Lempel compression algorithm is a string matching and parsing approach to data compression. The symbolwise equivalent for parsing models has been defined by Rissanen and Langdon and gives the same ideal codelength at the same cost in coding parameters. By describing the context and coding parameter for each symbol an insight is provided into how the Ziv-Lempel method achieves compression. This treatment does not employ a probabilistic source for the data string. The Ziv-Lempel method effectively counts symbol instances within parsed phrases. The coding parameter for each symbolwise context is determined by cumulative count ratios. The code string length increase for a symbolyfollowing substrings, under the symbolwise equivalent, is the log of the ratio of node counts in subtreessands\cdot yof the Ziv-Lempel parsing tree. To demonstrate the symbolwise equivalent of the Ziv-Lempel algorithm, we extend the work of Rissanen and Langdon to incomplete parse trees. The result requires the proper handling of the comma when one phrase is the prefix of another phrase.

A source code for binary strings, admitting a simple and fast hardware implementation, is described. The code is an arithmetic code, and it is capable of encoding strings modeled by stationary or nonstationary sources alike without use of alphabet extension. In particular, in the case with a stationary independent information source, the code degenerates to a bitwise implementation of Golomb's run-length code.

A new approach for black and white image compression is described, with which the eight CCITT test documents can be compressed in a lossless manner 20-30 percent better than with the best existing compression algorithms. The coding and the modeling aspects are treated separately. The key to these improvements is an efficient binary arithmetic code. The code is relatively simple to implement because it avoids the multiplication operation inherent in some earlier arithmetic codes. Arithmetic coding permits the compression of binary sequences where the statistics change on a bit-to-bit basis. Model statistics are studied from stationary, stationary adaptive, and nonstationary adaptive assumptions.

The problems arising in the modeling and coding of strings for compression purposes are discussed. The notion of an information source that simplifies and sharpens the traditional one is axiomatized, and adaptive and nonadaptive models are defined. With a measure of complexity assigned to the models, a fundamental theorem is proved which states that models that use any kind of alphabet extension are inferior to the best models using no alphabet extensions at all. A general class of so-called first-in first-out (FIFO) arithmetic codes is described which require no alphabet extension devices and which therefore can be used in conjunction with the best models. Because the coding parameters are the probabilities that define the model, their design is easy, and the application of the code is straightforward even with adaptively changing source models.

THIS issue of the IEEE TRANSACTIONS ON COMPUTERS iS devoted to the relatively new field of database machine design. What is a database machine? This question is answered in the context of the traditional approach to the storage, retrieval, and update of information. A collection of data is called a file, and the file is composed of individual records. The records themselves are partitioned into fields. In punched card technology, a file is a deck of cards, a record is a single card (a unit record), and a field is a grouping of adjacent columns to which the designer has assigned some meaning. Simple data processing systems access files in a sequential manner, processing each record in turn. This technique is well suited to the relatively inexpensive card and tape storage media. In the more expensive direct access (drum and disk) storage technologies, the storage media is divided into tracks. Tracks can be addressed directly, which leads to the concept of constructing an index of the file which indicates what data are stored on each track. With direct access storage devices and suitable file organization techniques, data processing programs can search, update, insert, and delete records from a file without processing the entire file in a sequential fashion. Direct access storage devices provide the user with more powerful functions, but these functions must be programmed on the central processor, as the IO interface traditionally communicates control information at a very low level.

Boolean functions can be realized using multiplexer elements by exploiting Shannon's expansion theorem. The cost of the realization often depends on the choice of the expansion variables versus the residue variables. A method employing Ashenhurst's decomposition charts [6] is described which assists in visualizing the residue functions. The method is effective for six variable functions or less.

Associative “logic-per-track” processors for data management are examined from a technological and engineering point of view. Architectural and design decisions are discussed. Some alternatives to the design of comparators, garbage collection, and domain extraction for architectures like the Relational Associative Processor (RAP) are offered.

In performing the operation of subtraction in additive systems, a popular practice is to complement the subtrahend and add. A second method of performing subtraction, which seems to have been overlooked, is to complement the minuend, add it to the subtrahend, and complement the result. In many cases, this second method is more awkward; however, in two instances it seems to be worthy of consideration. The first instance, surprisingly enough, concerns BCD systems, where minuend complementation can compare favorably with more conventional methods of BCD subtraction. The second instance concerns negative radix numbers where the technique of minuend complementation seems to offer definite advantages.

Armstrong, Friedman, and Menon have shown how fundamental-mode normal flow tables can be realized without delay elements under the constraint that line delays do not exceed a given chain of gate delays. The state assignments for these tables are designed to yield noncritical races and minimum transition times. This note presents an independently discovered alternative to their design procedure and also extends their results to totally sequential (or race-free) state assignments. The almost totally sequential state assignment is defined and presented as an approach toward delay-free realizations, under suitable line delay constraints, of nonnormal flow tables.

Abstract—Asynchronous circuits are analyzed from the stand- point of different constraints which may apply to the stray delays (zero-line delays, bounded delays, and the use of delay elements). The relationship between "hazard-free" sequential networks and "speed independent" circuits is illuminated. Speed independence is defined relative to a fundamental mode circuit (which in turn may be described by a flow table); the analysis is carried out from this stand- point. The allowed sequence graph may be used to detect critical races. A circuit is seen to be speed independent if and only if its circuit-derived excitation matrix has no 1-input change critical races. The generalization of the essential hazard has been studied in detail and is shown to be an obstacle to hazard-free realizations when line delays can be arbitrarily large.