scispace - formally typeset
Search or ask a question

Showing papers on "Run-length encoding published in 1997"


Patent
03 Apr 1997
TL;DR: In this paper, a subset of the recognized printable characters are designated as encoding characters for use in encoding data that includes characters not among the recognized printed characters, and each time that one of the encoding characters occurs in the input data, the byte representing that character must also be encoded.
Abstract: Characters that are not among those recognized for use in page messages are encoded using a subset of the recognized printable characters to enable the data comprising a page object to be transmitted over a paging channel. In one preferred form of the invention, a subset of the recognized printable characters are designated as encoding characters for use in encoding data that includes characters not among the recognized printable character set. Further, each time that one of the encoding characters occurs in the input data, the byte representing that character must also be encoded. To encode the characters, the byte is divided into nibbles. Each nibble is a hexadecimal digit that is encoded using one of the 16 encoding characters. If a byte of the input data repeats more than four times in succession, it is run length encoded (RLE) during the encoding process to compress the encoded data. Upon receipt, the encoded data are decoded using the same encoding characters, to recover the nibbles of each byte of the original input data. Furthermore, run length decoding is applied when a run length encoding signal character occurs in the encoded data stream. The run length encoding signal character is selected from the recognized printable character set. Using this technique, data can be sent that include characters not within the recognized printable character set and the encoded characters can be compressed to substantially reduce the amount of data transmitted as a paging object if the input data includes significant repeating characters.

11 citations


Proceedings Article
01 Jan 1997
TL;DR: A theoretical framework for -distortion limited compression that covers several recently proposed methods is presented and a comparison of coding results for the Lenna test image, a coronary angiogram, and a Landsat image is given.
Abstract: In -distortion limited compression each single pixel value is only changed by maximal grey values. In this paper we present a theoretical framework for -distortion limited compression that covers several recently proposed methods. The basics of each of these methods are described. We give a comparison of coding results for the Lenna test image, a coronary angiogram, and a Landsat image. Results are reported for various tolerances. Standard DPCM is used as a reference. While this paper gives an overview over various algorithms, the main purpose is to indicate what level of compression can be expected when limiting the error in -distortion sense. 1. A -DISTORTION LIMITED COMPRESSION FRAMEWORK In many applications, for example medical imagery, SAR imagery, or numerical weather simulations, the large amount of data to be stored or transmitted asks for data compression. Since lossless coding usually gives a compression ratio of at most 4:1, lossy coding methods have to be employed when higher compression ratios are needed. Most lossy compression schemes operate by minimizing some average error measure such as the root mean square error. However, in error critical applications such as medical imagery or target recognition, such average error measures are inappropriate. Instead, there is usually a need for a guarantee that a single pixel has not been changed by more than a certain tolerance (which may depend on the pixel location). Thus, the error in each pixel has to be controlled. In this paper we consider a -distortion limited compression scheme with global tolerance . For such an encoding method the code for a one-dimensional signal represents a reconstruction signal , where "! # %$ & !(') & +* -,/. & !0') & 1 24365 798;:= > @? $ ! 7 'A 7 $CB For D FE this leads to lossless compression. If is small, the term 'near-lossless coding' appears to be 0 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 9 10 Figure 1: Each left-to-right path is an element of GIH;J9KML with K/NOJQPSRTP6RVU6R=W"RXPSRXYZRV[6RX\SRTY6RT]ZL . justified. can be seen as the set of all left-toright paths in a trellis as depicted in Figure 1. Which of the _^ ` a X elements of b can be coded most efficiently? All coding methods described below use a lossless coding strategy c and try to determine, at least approximately or heuristically, the element of b that can be coded most efficiently using c . Mathematically, let d be the set of lossless coding methods. For example, ce fd could be a 0-order entropy coder. Then the coding problem for this particular c is: find !Vgh such that i c !Vg X j 24kml n 8"oqp6rmsut i c !X X wv where i Tx gives the length of the code or some estimate thereof. In the following sections we give short descriptions of several -based compression methods. Most of these methods were implemented and tested with the images given in Figure 2. 2. QUANTIZATION VS PRECONDITIONING In the problem formulation of the previous section the signal to be coded can be modified in each component independently of the other components. Thus, for a signal and y{z }| it is possible that yT ~  |€ but yT z  |; . In other words, is in general not simply the result of using a quantizer for the values of . We refer to as a preconditioned version of in contrast to a quantized version. The emphasis in this paper is on preconditioning. Nevertheless, the problem of finding the quantization function such that the quantized version of has minimal 0-order entropy can be solved in time using dynamic programming [1]. It is also shown that for a tolerance E the entropy savings are at most _^ ` a per pixel. 3. ENTROPY-CODED DPCM BASED METHODS The entropy coding of the prediction residuals of a DPCM scheme is a standard method for lossless compression. It can easily be modified to serve as an distortion based compression method. DPCM1. The signal is uniformly quantized with quantization bin size ^ `#a . Thus, a quantized version of the original signal is computed. Then the residuals of a linear predictor are entropy coded. The disadvantage of this method is that for larger there are only a few different grey levels leading to 'plateau effects'. DPCM2. No a priori grey value reduction is performed, but the prediction error of the DPCM scheme is uniformly quantized to match the desired tolerance . When the predictor coefficients are not integer values, this method does not coincide with the method DPCM1 and does not show the plateau effects. Results for several medical images are reported in [2]. In the above mentioned methods, there is actually no mechanism to minimize the entropy of the error sequence. When we use a lossless predictive coder followed by an entropy coder, the optimization problem wv asks for the path in the trellis whose corresponding residual sequence has minimum entropy. We conjecture that this optimization problem is NP-hard. Note that the complexity depends on the signal length and the tolerance . We applied genetic algorithms (GA) [3, 4] to solve this optimization problem for a signal and a tolerance . GA. In our setting a chromosome is a word of length over the alphabet @' . x"x x . E . x x x .X -, and represents the signal ` . The genetic operations are 2-point crossover and mutation. We use roulette wheel parent selection. The evaluation of a chromosome is given by the entropy of the distribution of prediction residuals of { 1 ` . For the fitness function we use exponential ranking. Large tests for the determination of suitable parameters were performed. The results obtained with the GA approach are rather disappointing. For example, as a signal a line of the image Lenna was taken. The entropy of that signal is 7.0, after quantization with tolerance ^ , and after prediction the sequence can be coded with an entropy of 3.1. The solution found with the GA only gave an entropy of 3.9. Thus, the GA is not even able to beat the method DPCM1. The minimum-entropy constrained-error DPCM (MECE) of [5] is another method that tries to minimizes the entropy of the prediction residual sequence. It uses an iterative optimization method that arrives at a local optimum. MECE. Assume that an ideal entropy coder is given for a fixed residual distribution. To find the optimal element of for this coder one has to solve a shortest path problem. This can easily be done via Dynamic Programming. Now, using an entropy coder that is optimal for the actual residual distribution will give a decrease in entropy. These two steps are performed iteratively until a stopping criterion is matched. For images a two dimensional 3-tap predictor is used and the images are coded row by row. The results can be further improved by using a 1-order entropy coder with a certain number of contexts. In the above mentioned methods the predictor and the contexts are fixed. Of course, it would be advantageous to include the choice of predictor coefficients and context into the optimization problem; clearly, this makes the problem even more complicated. A sophisticated method that uses adaptive context modeling to correct prediction biases is the -constrained CALIC [6]. The converse problem of determining a predictor such that the prediction residuals have minimum entropy was investigated for lossless coding in [7]. 4. PIECEWISE LINEAR CODING Piecewise linear coding (PCL) is a generalization of Run Length Encoding. It is also called fan-based coding; for an extensive overview see [8]. In piecewise linear coding a signal is split into segments each of which can be described by a linear function. Each segment then is coded by the length of the segment and the slope parameter. The constant additive part of the function is implicitly given by the previous segment; only for the first segment the initial signal value has to be coded. For example, the signal in Figure 1 is represented as 3(1,0)(2,-1)(3,2)(3,-1). In the case that i Tx counts the number of segments the optimization wv can be solved in -time via Dynamic Programming [9]. In [10, 11] a suboptimal greedy method that works in linear time is proposed for the same optimization problem. Essentially, it works as follows. The image is transformed into a 1-dimensional signal, e.g., by a Hilbert-Peano scan. Then the linear segments are successively determined: starting at the endpoint of the last determined segment, the new segment is chosen to be the one of greatest possible length. Finally, an 0-order entropy coder is applied to the list of segment lengths and segment slopes. Better results can be obtained when the length of the 0-order entropy code is minimized in place of Figure 2: The a ^ a ^ 8-bit test images Lenna, Angio, Landsat.

10 citations


Journal ArticleDOI
TL;DR: Good performance is proved in two different applications: the generation of homotopic skeletons through thinning processes, and the extraction of linear features through serializing multiangle parallelism operations.
Abstract: Image compression techniques have been recently used not only for reducing storage requirements, but also computational costs when processing images on low cost computers. This approach might be also of interest for processing large engineering drawings, where feature extraction techniques must be intensively applied for their segmentation into regions of interest for subsequent analysis. This paper explores this alternative using a simple run-length compression, leading to excellent results. Although this approach is not new and can be classified within the decomposition paradigm used since the early stages of line drawing image processing, the developed formalism allows directional morphological set transformations to be performed, on a low cost personal computer, faster than on costly parallel computers for the same, but uncompressed, images. This good performance is proved in two different applications: the generation of homotopic skeletons through thinning processes, and the extraction of linear features through serializing multiangle parallelism operations.

4 citations


Patent
07 Mar 1997
TL;DR: In this article, the problem of improving compressibility and reducing the time of a restoration processing in the compression technique of picture signals using run length encoding is addressed by using a run length calculation.
Abstract: PROBLEM TO BE SOLVED: To improve a compressibility and to shorten the time of a restoration processing in the compression technique of picture signals using run length encoding. SOLUTION: At the time of one-dimensionally encoding and compressing the picture signals, a white (or black) run length at present calculated in a run length calculation part 1 is compared with the white (or black) run length stored one cycle earlier in a previous data storage part 2 in a comparison part 3. When they do not match, a code chart 6 prepared beforehand is referred to and conversion to codes is performed in a code conversion part 4. When they match, the conversion to prescribed repetitive codes is performed in a repetitive code conversion part 5 without referring to the code chart 6.

3 citations


Patent
21 Apr 1997
TL;DR: In this article, a line drawing compressed by run length encoding is read out, line by line, run data connecting to previous lines are investigated and divided into blocks of running data connecting with previous lines one to one, and connection information between the run data divided into the blocks is added.
Abstract: PROBLEM TO BE SOLVED: To easily extract outline data from line drawing data (binary image data). SOLUTION: Run data 92 having a line drawing compressed by run length encoding are read out, line by line, run data connecting to previous lines are investigated and divided into blocks of run data connecting to previous lines one to one, and connection information between the run data divided into the blocks is added, so that the data are managed as run block data 93 to 98. The connection information between the mutually connecting run block data is traced downward for the left-end coordinates of the run block data 93 to 98 and upward for the right-end coordinates to extract coordinate data (outline data) 120 and 121. COPYRIGHT: (C)1998,JPO

2 citations


Patent
19 Aug 1997
TL;DR: In this paper, a decoding part 1 Huffman decodes and run length decodes the compression data of one block and generates quantized data Ruv form the matrix of 8×8 for instance.
Abstract: PROBLEM TO BE SOLVED: To effectively convert images by reducing the amount of data to be prepared in the case of performing expansion to the images of various sizes based on the compression data of the images and shortening the processing time for inverse quantization in the case of generating reduction images. SOLUTION: A decoding part 1 Huffman decodes and run length decodes the compression data of one block and generates quantized data Ruv. The quantized data Ruv form the matrix of 8×8 for instance. Since Huffman encoding and run length encoding are reversible encoding, the decoded quantized data Ruv are completely the same as the quantized data Ruu at the time of performing JPEG(joint photographic expert group) compression. A reduction part 2 omits the high frequency component of the quantized data Ruv and performs conversion to the matrix of a small dimension. Thus, the amount of the data to be prepared is reduced in the case of performing the expansion to the images of the various sizes based on the compression data of the images. Then, in the case of generating the reduction images, the processing time for the inverse quantization is shortened and the reduction images faithful to source images are obtained.

2 citations


Proceedings ArticleDOI
02 Nov 1997
TL;DR: Efficient lossless compression methods applicable to mixed types of data, in which there is a data base containing files with signal and text data, and the files are accessed via a narrow-band channel such as a modem are described.
Abstract: We describe efficient lossless compression methods applicable to mixed types of data. The methods are applicable to situations in which there is a data base containing files with signal and text data, and the files are accessed via a narrow-band channel such as a modem. Theoretical principles are presented, including a discussion of measures of compressibility. Experimental results are also presented.

1 citations


Proceedings ArticleDOI
01 Sep 1997
TL;DR: This paper presents a new algorithm for image compression which makes use of a variation of the run-length encoding and Huffman codes and is almost 53% more efficient than the version of the Lempel-Ziv-Welch algorithm used in GIF files.
Abstract: This paper presents a new algorithm for image compression which makes use of a variation of the run-length encoding and Huffman codes. This algorithm was tested on monochromatic images of letters and documents. A comparative study with other compression methods is presented. In the best case, our algorithm is almost 53% more efficient than the version of the Lempel-Ziv-Welch algorithm used in GIF files.