1、STD*AIIM TR33-ENGL 1998 E 1012348 0501784 O22 ASSOCIATION FOR INFORMATION AND IMAGE MANAGEMENT INTERNATIONAL 1100 Wayne Avenue Suite I100 Silver Spring, Maryland 20910 AllM 301.587-8202 STD-AIIM TR33-ENGL 3998 3032348 0503785 Th9 m ANSVAIIM TR33-1998 0 by the Association for information and Image Ma
2、nagement International 1 100 Wayne Avenue, Suite 1 100 Silver Spring, MD 20910-5603 Tel: 3011587-8202 e-mail: aiimai.im.org Website: http:/www.aiim.org Fax: 301/587-2711 ISBN 0-89258-349-5 Prinkd in the United States of America STD-AIIM TR33-ENGL 1978 1012348 0501786 7T5 H ANSUAIIM TR33-1998 Technic
3、al Report for Information and Image Management - Selecting an Appropriate Image Compression Method to Match User Requirements An ANSI Technical Report prepared by the Association for Information and Image Management International Abstract: This technical report provides practical methods for analyzi
4、ng user requirements for image compression in order to select an appropriate and optimal image compression scheme that matches user requirements. The report describes procedures for applying image compression methods and standards to various types of documents in Electronic Image Management (EM) sys
5、tems. it addresses some of the applications of image compression methods and standards that are used for continuous tone (color and gray scale) images and for bi-level (line art and text) images, and it provides guidance in selecting applicable compression algorithms for each among a wide range of s
6、ource documents. ANSYAIIM TR33-1998 Selecting an ApprOpnate Image Compression Method to Match User Requirements Contents 1 Purpose I P Foreword i 2 Normative references . 1 3 Definitions 1 4 Abbreviations . 3 5 Executive summary 3 6 Description of a document 5 7 Compression categon es . 10 8 Image p
7、resentation . 11 9 Choose a compression methodology 12 1 o Conclusions 12 Table Table 1 Questionnaire . 12 Foreword (This foreword is not part of the American National Standard for Information and Image Management TR33-1998 - Technical Report - Selecting an Appropriate Image Compression Method to Ma
8、tch User Requirements.) This report describes procedures for applying image compression methods and standards to various types of documents in Electronic Image Management (EM) systems. It addresses some of the applications of image compression methods and standards that are used for continuous tone
9、(color and gray scale) images and for bi-level (line art and text) images. Standards that are discussed include ISOIIEC 11514, known as JBIG (Joint Bi-level Imaging Group), ISOIIEC 10918. known as JPEG (Joint Photographic Expert Group), and the IW-T Rec. T.4 rGroup 3”) and T.6 CGroup 43 compression
10、schemes. This report does not attempt to address all compression schemes, but focuses only on those that am current national or international standards in the area of document image management A number of international standards are in development which may be applicable in the future. It is planned
11、 that this document will be revised as newer compression schemes become available and standardized. P This technical report is intended to cover scanned documents that are to be compressed, processed, andior stored in an EIM system. Several strategies are described, including the use of single and m
12、ultiple compression schemes for image storage or processing. Various parts of this document should be applicable to a wide fange of systems. r This document is intended for use by managers and implementers involved in the acquisition and development of EIM systems. Suggestions for improving this tec
13、hnical report are welcome. They should be sent to the Chair, AIM Standards Board, Association for Information and Image Management Intemational, 1 100 Wayne Avenue, Suite 1100, Silver Spring, Matyiand, 2091 0-5603. At the time it approved this technical report, the AIIM Standards Board had the follo
14、wing members: Name of Oraanizaion ReDresented Representative Marilyn Wright, Chair Association for Infomiation and Image Management International Robert Breslawski Eastman Kodak Company Jewel M. Draas Bell i% Howell, Pub. Software Div. Basil Manns Library of Congress Roy Pierce Xerox Corporation Fer
15、nando L. Podio National Inst. of Stds either text or halftones of much higher contrast 3. Fingerprint photographs 5 Association for nfrmation and image Management hiemational STDeAIIM TR33-ENGL 3978 M 3032348 050L775 708 W ANSYAIIM TR33-1998 Selecting an Appropriate image Compression Method to Match
16、 User Requirements 9 Micrographics i. High contrast microfonns (e.g., document imaging, computer output microforms (COM) 2. Low contrast microfoms (e-g., X-ray microfonns, archiving systems) Documents with mixed content are perhaps the most common in actual practice. These may include: maps and char
17、ts (Maps and charts someimes contain only text and line art, but mosi often they also contain color solid tones and halftones.) general forms (These are usually text with line art, lines, and hamone shaded areas. General foms contain various colors and densities of handprinted characters.) - remitta
18、nce forms (These are usually light color text with halflone shaded areas. These forms contain various color and densty handpnnted characters.) - checks (Checks are usually text with halftone areas. They contain various colon and density handpnnted characten.) - charts (Charts are output of recorders
19、 with dark lines, or data, on recording paper with lighter color grids.) - serials (Serials are magazine artides containing text with halftones that may be monochromatic or dor.) - compound documents (Compound documents contain more than one content type, for example character text and images.) 6.2
20、Document contents Image fidelity and the intended usage pattem for the image represent two of the factors necessary to develop an image processing scheme. The third fador - document type - requires the examination of typical and atypical documents. A complex document is generally composed of many di
21、fferent types of content regions. For instance, part of the document migM be text that will be extracted and used to control retrieval of the document. Another part of the document might be line drawings, halftone images, of photographs. Each of these has charaderistics that differ widely and couid
22、be optimally compressed using different algorithms. It is necessary to examine a representative sampling of documents to determine what types of graphical content regions are present. Content types can be categorized: - continuous tone regions - halffone regions - line art and text regions The conte
23、nt type that exists in specific regions of a document image is different from the method of capturing the image. For instance, it is typical to capture black and white text images using a continuous tone scanner. Within the scanner, the continuous tone image is thresholded into a suitable binary ima
24、ge. In that case, the capture method is continuous tone while the content type is line art and text regions. The representation used for these bimodal content types is a binary image. Depending on the specific document that is examined, different content types can exist. Many of the compression algo
25、rithms can be applied to multiple content types. For example, both halfone and text regions might be compressed BS binary images using a binary compression technique. The JBIG algorithm wo- well for halftone images, line, and text images. Alternatively, the binary image of the halftone region can be
26、 converted to a continuous tone image and the document compressed using two algorithms - a continuous tone algorithm for the photographic content and a bimodal algorithm for the text content. 6.2.1 Continuous tone images A continuous tone image is characterized by a luminance scale that is continuou
27、sly variable from dark to light. A digital image is limited to a finite number of intervals. For instance, 8 bit gray represents 256 levels from O to 255. Four bit gray represents 16 levels from O to 15. A typical example of a continuous tone image is a photograph. Dithering is a technique used in s
28、canning and display to represent one color or tone by a pattern of other colon or tones. For example, a continuous shade of gray can be represented by a pattern of black and white pixels, using a higher portion of Mack pixels for the darlier shades of gray. A color that is not available can be repre
29、sented by aiiemating pixels of other colors - for example, a shade of pink may be represented by various numbers of red and white pixels. Some scanners have a feature to detect continuous tone areas, and automatically dither that portion of the page - presenng a series of ones and terns representing
30、 a pattem of black and white pixels (assuming the document is limited to black and white without gray) io approximate the gray in that area. The tesuking image is much more attractive than an all black or all white“ approximation of the continuous tone 1 1 6 Association far Informaton and lmage Mana
31、gement hkrriational STD-AIIM TR33-ENGL L99A E LOL2348 0501796 844 .I ANSVAIIM TR33-1998 Selecting an Apptopriate image Compression Method to Match User Requirements area, but is much less efficient to compress. If the continuous tone portion is important, it is much more appropriate to use a scannin
32、g and storage technique designed for continuous tone Images, such as JPEO. r Fmm a compression point of view, it is more desirable to use a lossy JPEG compression than a lossless compression method, because the lossy method provides a greater amount of compression. Given the class of document, a num
33、ber of options are available. By empirical testing, the options that are best for the specific case can be chosen. Merver the intended use of the image is critical, such as in the case of fingerprints, medical images, or legally qualified evidence, it will be necessary to qualify the use of a lossy
34、compression. This is especially true if morphological or image enhancement techniques are used to improve the viewing of the resultant image. The selection of quantization tables for JPEG greatly affects this process. The example quantization tables3 are based on psychovisual testing with photograph
35、s. 6.2.2 Haiftone images To accommodate certain printing processes, the appearance of a continuous tone image is simulated by a halftone. which is a pattern of dots using a limited number of colors of variable sites or variable spacing. For instance, a black and white photograph in a newspaper Is cr
36、eated using just black on a white background. The percentage of the area covered by the dots in a given section simulates a level of gray. The output of a color inkjet printer can give the appearance of many colors using a pattern of three or four colors. P - Halftone images can be converted into co
37、ntinuous tone images. 6.2.3 Line art and text images Line art or text images represent line drawings or textual content regions. They are typicaly binary images (single bit black and white). Examples of where they are used to represent multi-modal content Include: - wiring diagrams - business charts
38、 - documents printed using highlight colors In these cases, each color is represented as a separate component and colors are not mixed to get new colors. A muiti-modal content region typically comprises text, charts, or line drawings. In a pure sense, these represent discrete values (black, white, r
39、ed, light green, dah blue) rather than a continuously variable color (represented using RGB or CMYK-tuples). The methods used to compress this component are well understood and include JBIG, T.4, and T.6. It is generally optimal to split the image into multiple components, each representing a separa
40、te color plane, and to compress the components separately using a combination of JPEG and T.4lT.6 or JUG. The process of splitting the image into these components is straightforward and is available in many scanning and editing pmgrams. In some cases, it is not possible to split the dowment into the
41、se components. Then the choice is to select the best single cornpression method that performs the muiti-modal components adequately. At a distance, halftone images are interpolated by the eyes to appear as a continuous tone image. A similar mechanism can be applied in a specially designed scanner to
42、 allow a halftone image to be interpolated into a continuous tone Image. Haiftone images are discussed separately from line 6.3 Image usage The compression seledion process must take into account a large number of options and choices. Important factors to be considered are the intended uses of the i
43、mage. art or text images for two reasons: - The assumptions behind compression algorithms that work well for halftone images, such as JBIG, are different from the assumptions underiying compression algorithms that only apply to text and line art binary data (3.4, T.6). Generally, halftone compressio
44、n algorithms work well for either data. It is best to split the decision process into discrete steps. nie document must be examined and classified along with the intended usage of the information extracted from the doaiment. Some applications have multiple needs. For instance, the images might be ca
45、ptured and stored in a database. The data contained in the images might then e extraded and used for referen infomation. The combination of the two might be used to prepare information for distribution. One example of such an application would be an online encyclopedia. %he example quantization tabl
46、es in the PEG standard arc often used as the default in JPEG implementations and have become known as defuulr tables. They were not originally intended as such by the PEG committee. e 7 Association far information and Image Uanagement nkmationai In a work flow process, the document being captured ca
47、n be retained at the highest level of quality for archiving and infrequent or scholarly reference. A version of the same document can be produced at a much lower quality and widely distributed. The image quality must be the best at the point of capture and during the initial processing. Any compress
48、ion applied at that point should minimize loss. A compression scheme can amplify deficiencies in the scanning process. The best example is a half-tone pattem that is scanned into the system. The scanner adds sampling error becauce of the difference in alignment and resolution of lhe haitlone pattern
49、 and that of the sensor amy in the scanner. On a continuous tone scanner, the halftone patiem is distotted and instead of being bimodal lack and white), it becomes a gray toned pattem that is a distortion of the original halftone. On a binary scanner, the pattem can produce visible artifacts. The compression of halftones is a specific problem and some compression standards, such as JBIG, are designed to address this directly. Since JBIG can be lossless, none of the captured data is lost. During display or printing, the distorted halftones must be processed and provided in
