Introduction
Images are an important medium from which human beings observe the majority of the information they received from the real world. In its general sense, the word “image” could include all entities that can be visualized, such as a still image, video, animation, graphics, charts, drawings, even also text, and so forth. Nowadays, “image” rather than “picture” is used because computers store numerical images of a picture or scene. Image techniques, which are expanding over wider and wider application areas, have attracted more and more attention in recent years. Image engineering (IE), an integrated discipline/subject comprising the study of all the different branches of image techniques, is evolving quickly.
From 1969 to 2000, a well-known bibliography series had been developed to offer a convenient compendium of the research in picture processing until 1986, as well as in image processing and computer vision after 1986. This series has been ended in 2000 by the author after a total of 30 survey papers were published (Rosenfeld, 2000a). Some limitations of this series for the termination are (Zhang, 2002b):
1. No attempt is made to summarize the cited references for each year.
2. No attempt is made to analyze the distributions of the selected references from various sources.
3. No attempt is made to provide statistics about the classified references in each group.
Another survey series, but on IE, has been started since 1996 (Zhang, 1996a, 1996b, 1996c, 1997, 1998, 1999, 2000a, 2001a, 2002a, 2003, 2004, 2005). The purpose of this survey work is mainly to capture the up-to-date development of IE, to make available a convenient means of literature searching facility for readers working in related areas, and to supply a useful reference for the editors of journals and potential authors of papers. This new series overcame the weakness of the earlier mentioned one by summarizing the cited references for each year, analyzing the distributions of the selected references from various sources, and providing various statistics about the classified references in each group. This new survey series has already made consecutively for ten years. This article will present an overview of this survey series by showing the idea behind and consideration on this work as well as the comprehensive statistics obtained from this work.
background
Image Engineering
IE, from a perspective more oriented to technique, could be referred to as the collection of three related and partially overlapped groups of image techniques, that is, image processing (IP), image analysis (IA), and image understanding (IU). In a structural sense, IP, IA, and IU build up three interconnected layers of IE as shown in Figure 1. Each of them operates on different elements (IP’s operand is pixel, IA’s operand is object, and IU’s operand is symbol) and works with altered semantic levels (from low at IP to high at IU).
Figure 1. Three layers of image engineering
The three layers follow a progression of increasing abstract-ness and of decreasing compactness from IP to IU.
IP primarily includes the acquisition, representation, compression, enhancement, restoration, and reconstruction of images. While IP is concerned with the manipulation of an image to produce another (improved) image, IA is concerned with the extraction of information from an image. Compared to IP which takes an image as input and outputs also images, IA takes also an image as input but outputs data. Here, the extracted data can be the measurement results associated with specific image properties or the representative symbols of certain object attributes. Based on IA, IU refers to a body of knowledge used in transforming this extracted data into certain commonly understood descriptions and for making subsequent decisions and actions according to the interpretation of the images.
Related Subjects
IE is a broad subject encompassing studies of mathematics, physics, biology, physiology, psychology, electrical engineering, computer science, automation, and so forth. Its advances are closely related to the development of telecommunications, biomedical engineering, remote sensing, document processing, industrial applications, etc. (Zhang, 2002b).
According to different science politics/perspectives, various terms such as computer graphics (CG), pattern recognition (PR), computer vision (CV), scene analysis (SA) (just counted as another name of CV, see Rosenfeld, 2001) etc., are (partially) overlapped with IP, IA, and/or IU. A diagram describing the relationship among the earlier-mentioned subjects is given in Figure 2. Images are captured from the real world and processed to furnish the basis for IA or PR. The former produces data that can be visualized by CG techniques while the latter continually classifies them into one of several categories. Results produced by both of them can be further interpreted for human beings to understand the real world. The whole process aims to make computers capable of understanding environments from visual information, which is also the purpose of CV/SA.
THE CURRENT “PICTURE” OF IMAGE ENGINEERING
What is the current “picture” of IE? Answering this question is the foremost intention of the new survey series. For such a purpose, selection of reference source and classification of references according to contents are two important factors. Also for such a purpose, three statistics made by this survey are illustrated in the following.
Classification Scheme
The classification scheme used in the bibliography series should reflect the contents of references. A classification problem can be considered as a problem of partitioning a set into subsets. An appropriate classification of references into groups and/or sub-groups should satisfy the following four conditions:
1. Every reference must be in a group.
2. All groups together could include all references.
3. The references in the same group should have some common properties.
4. The references in different groups should have certain distinguishing properties.
Taking into consideration these conditions and the status of development in the field, a complete and compact classification of the theories and techniques of IE is proposed and listed in Table 1 (Zhang, 2002b). It is easy to verify that these conditions are fulfilled by this classification.
Figure 2. Image engineering and related subjects
Table 1. Classification scheme of image engineering
Group | Sub-group | |
P1 | Image capturing and storage (including camera calibration) | |
P2 | Image reconstruction from projections | |
IP: Image Processing | P3 | Filtering, transformation, enhancement, restoration |
P4 | Image and/or video coding and standards | |
P5 | Image digital watermarking and image information hiding | |
A1 | : Edge detection, image segmentation | |
A2 | : Representation, description, measurement (bi-level image) | |
IA: Image Analysis | A3 | Analysis of color, shape, texture, position, motion, etc. |
A4 | : (2-D) object recognition, extraction, tracking, classification | |
A5 | Human face and organ detection and location | |
U1 | : (Sequential, Volumetric) image registration and matching | |
IU: Image Understanding | U2 | 3-D modeling, representation and real world recovery |
U3 | : Image interpretation and reasoning (semantic, expert system) | |
U4 | Content-based image and video retrieval | |
T1 | System and hardware (fast algorithm implementation) | |
T2 | Telecommunication, television | |
TA: Technique Applications | T3 | Documents (texts, digits, symbols) |
T4 | Bio-medical imaging | |
T5 | Remote sensing, surveying and mapping | |
T6 | Others |
Table 2. Selected journals and their abbreviations
# | Journal | Abbreviation. |
1 | Acta Automatica Sinica | AAS |
2 | Acta Electronica Sinica | AES |
3 | Acta Geodactica et Cartographica Sinica | AGCS |
4 | Chinese Journal of Biomedical Engineering | CJBE |
5 | Chinese Journal of Computers | CJC |
6 | Chinese Journal of Stereology and Image Analysis | CJSIA |
7 | Computerized Tomography Theory and Applications | CTTA |
8 | Journal of China Institute of Communications | JCIC |
9 | Journal of Data Acquisition and Processing | JDAP |
10 | Journal of Electronic Measurement and Instrument | JEMI |
11 | Journal of Electronics and Information | JEI |
12 | Journal of Image and Graphics | JIG |
13 | Journal of Remote Sensing | JRS |
14 | Pattern Recognition and Artificial Intelligence | PRAI |
15 | Signal Processing | SP |
Source Selection
As with any other emerging discipline, a large number of references related to IE have been published worldwide. The continued growth of the literature has already made it impractical to cover all of them in one survey (Rosenfeld, 1999). Though references have been dispersed across many resources, the most popular ones are conference proceedings, journals, and topics. Considering the fast publishing rate, the conference proceedings would be ranked first followed by journals and topics. Considering the comprehensiveness, the topics would be ranked first followed by journals and conference proceedings. Considering the quality and coverage, journal articles would be ranked higher than that of conference proceedings and topics. Combining all these considerations, journals would be the best choice for such a survey series.
Based on a careful selection of literature for providing an appropriate coverage in this area, 15 importantj ournals (in the sense defined by Lin & Zhang, 1996) with high standard articles that are published in Chinese have been selected to limit the volume of references to a manageable size. All of the papers in these journals have titles, abstracts, and keywords in English. The list of journals is given in Table 2.
Summary over Years
The first statistic made from this survey is a summary of the number of publications in the last ten years, as shown in Table 3. As in a survey of papers, the references have been classified into five groups: IP, IA, IU, TA and Survey. In Table 3, the total number of papers published in the selected journals (#T), the number of papers selected for survey as they are related to IE (#S), and the selection ratio (SR), which equals to #S/#T, for each year have been provided. In addition, the paper numbers for five groups (and their percentages in the year) are also listed.
Some interesting points can be noted from Table 3:
Table 3. Summary over the last 10 years
Year | #T | #S | SR | IP | IA | IU | TA | Survey |
1995 | 997 | 147 | 14.74 | 35(23.8%) | 52(35.4%) | 14(9.52%) | 46(31.3%) | |
1996 | 1205 | 212 | 17.59 | 52(24.5%) | 72(34.0%) | 30(14.2%) | 55(25.9%) | 3(1.42%) |
1997 | 1438 | 280 | 19.47 | 104(37.1%) | 76(27.1%) | 36(12.9%) | 60(21.4%) | 4(1.43%) |
1998 | 1477 | 306 | 20.72 | 108(35.3%) | 96(31.4%) | 28(9.15%) | 71(23.2%) | 3(0.98%) |
1999 | 2048 | 388 | 18.95 | 132(34.0%) | 137(35.3%) | 42(10.8%) | 73(18.8%) | 4(1.03%) |
2000 | 2117 | 464 | 21.92 | 165(35.6%) | 122(26.3%) | 68(14.7%) | 103(22.2%) | 6(1.29%) |
2001 | 2297 | 481 | 20.94 | 161(33.5%) | 123(25.6%) | 78(16.2%) | 115(23.9%) | 4(0.83%) |
2002 | 2426 | 545 | 22.46 | 178(32.7%) | 150(27.5%) | 77(14.3%) | 135(24.8%) | 5(0.92%) |
2003 | 2341 | 577 | 24.65 | 194(33.6%) | 153(26.5%) | 104(18.0%) | 119(20.6%) | 7(1.21%) |
2004 | 2473 | 632 | 25.60 | 235(37.2%) | 176(27.8%) | 76(12.0%) | 142(22.5%) | 3(0.47%) |
Total | 18819 | 4032 | 1364(33.8%) | 1157(28.7%) | 553(13.7%) | 919(22.8%) | 39(9.67%) | |
Average | 1882 | 403 | 21.44 | 136.4 | 115.7 | 55.3 | 91.9 | 3.9 |
Figure 3. Number variation offour groups in selected publications for last 10 years
1. This work has a quite large scale with nearly 19,000 papers involved and more than 4,000 papers selected and classified.
2. IE is an important topic for electronic engineering, computer science, and automation. The average SR is more than 1/5, which is remarkable considering the wide coverage of these journals.
3. IE publication evolves quite steadily. From Table 3, #S is increasing every year, and its value in 2004 is four times bigger than 10 years’ ago. It is also noted that SRs in the recent three years are not only rising but also among the highest in ten years with SR > 1/4 for 2004.
4. The number of publications for IP and IA constitute 2/3 of the total number of IE publications. This shows the current research focus of IE. In contrast, research work on IU needs to be promoted.
5. The growing rates of publications for IP, IA, IU, and TA are comparable. To make it clear, Figure 3 shows the numbers of publications for these four groups graphically. The four curves in Figure 1 run quite smoothly and have not intercrossed in the last five years.
Distribution Analysis
The second statistic is the summary over the different journals (see Table 2), and the results are shown in Table 4. In Table 4, #I is the number of surveyed issues; #T and #S are now the total number of papers and the number of survey-selected papers, respectively. We also give the rank of the different journals according to their SR (selection ratio), and the rank of the different journals according to TR (total ratio, i.e., over all 15 journals). In Table 4, SR gives the relative frequency of IE publications in a journal. This relative frequency brings a measure of the probability of obtaining useful information from thatj ournal. TR presents the relative contribution of each journal to IE publication and supplies a figure of importance of that journal among 15 journals. According to these rankings, readers could selectively scan the journal and judge the value of each journal.
From Table 4, the following observations can be made:
1. SR of a journal gives the probability of obtaining useful information from this journal. JIG has the highest SR among the 15 journals, and therefore, it should be checked frequently.
2. TR of a journal shows the contribution of this journal to IE publication. JIG has the highest TR among the 15 journals (and much higher than all competitors); therefore, it is evident that this journal offers a focused location for researchers in this field.
3. According to the scatter rule (Ding, 1993), most research papers of one discipline will be concentrated in a few number of journals, and other papers will be dispersed in a large number of journals. The leading fivejournals: JIG,AES, CJC, PRAI, and JEI, contained more than twice the number of IE papers compared to the other 10 journals.
Table 4. Summary over 15 journals
Journal | #I | #T | #S | SR (Rank) | TR (Rank) |
AAS | 60 | 1280 | 132 | 10.31% (14) | 3.27% (10) |
AES | 88 | 3474 | 504 | 14.51% (11) | 12.5% (2) |
AGCS | 40 | 592 | 95 | 16.04% (9) | 2.36% (13) |
CJBE | 48 | 822 | 121 | 14.72% (10) | 3.00% (11) |
CJC | 120 | 1946 | 319 | 16.39% (7) | 7.91% (3) |
CJSIA | 36 | 488 | 111 | 22.75% (5) | 2.75% (12) |
CTTA | 40 | 460 | 74 | 16.09% (8) | 1.84% (14) |
JCIC | 106 | 2017 | 209 | 10.36% (13) | 5.18% (8) |
JDAP | 40 | 971 | 211 | 21.73% (6) | 5.23% (7) |
JEI | 84 | 2095 | 284 | 13.56% (12) | 7.04% (5) |
JEMI | 40 | 565 | 57 | 10.09% (15) | 1.41% (15) |
JIG | 102 | 1704 | 1223 | 71.77% (1) | 30.3% (1) |
JRS | 48 | 639 | 155 | 24.26% (4) | 3.84% (9) |
PRAI | 41 | 823 | 291 | 35.36% (2) | 7.22% (4) |
SP | 48 | 943 | 246 | 26.09% (3) | 6.10% (6) |
Summary | 941 | 18819 | 4032 |
Detailed Classification Statistics
The third statistic is a detailed classification of IE publications in each sub-group and for each journal. The results are listed in Table 5.
Many commentaries could be made on Table 5; however, we only point out three important observations:
1. From the number of publications in different subgroups, it seems that image compression (P4), image segmentation (A1), and object extraction (A4) are the most important research topics in all these years. Note that classes A1 and A4 are closely related but different. A1 is concentrated for separating an image into meaningful parts while A4 is more related to direct detection with object model; the former is more like an unsupervised task and the latter is more supervised.
2. The detailed classification shows that differentj ournals have their different emphasis; some of them cover different sub-groups of IE (for example, AES, CJC, JDIP, JEI, JIG, PRAI) evenly, while some of them are more specialized in certain sub-groups of IE (for example, CJBE for T4, CTTA for P2, JCIC for P4 and T2, and JRS for T5). That information would be useful for potential authors.
3. The top two sub-groups are P4 and A1, respectively. Both of them contain about 1/8 of the total publications and thus indicate two hot research areas in IE. How ever, a detailed comparison for each year, as shown in Figure 4, illustrates that they have had quite different developing trends. Image coding had been progressed mostly from 1997 to 2001 and decreased since than. This can be seen clearly by the polynomial (IC) curve (which is the polynomial approximation of the image coding curve for the last 10 years) in Figure 4. On the other side, image segmentation is progressing steadily all these years.
Table 5. Detailed classification of references
Journal | P1 | P2 | P3 | P4 | P5 | A1 | A2 | A3 | A4 | A5 | U1 | U2 | U3 | U4 | T1 | T2 | T3 | T4 | T5 | T6 | S1 |
AAS | 16 | 7 | 7 | 5 | 17 | 5 | 1 | 15 | 11 | 10 | 17 | 3 | 3 | 1 | 7 | 1 | 6 | ||||
AES | 22 | 16 | 40 | 97 | 41 | 69 | 16 | 14 | 24 | 21 | 36 | 9 | 3 | 17 | 25 | 22 | 15 | 6 | 7 | 4 | |
AGCS | 10 | 6 | 5 | 4 | 6 | 9 | 7 | 14 | 3 | 3 | 1 | 20 | 6 | 1 | |||||||
CJBE | 6 | 14 | 5 | 2 | 13 | 1 | 1 | 2 | 8 | 6 | 5 | 58 | |||||||||
CJC | 22 | 4 | 17 | 32 | 23 | 40 | 21 | 9 | 16 | 23 | 21 | 21 | 7 | 17 | 5 | 6 | 23 | 3 | 2 | 6 | 1 |
CJSIA | 4 | 4 | 7 | 5 | 3 | 16 | 12 | 4 | 5 | 1 | 6 | 2 | 6 | 1 | 20 | 2 | 11 | 2 | |||
CTTA | 7 | 42 | 3 | 1 | 1 | 1 | 1 | 2 | 9 | 7 | |||||||||||
JCIC | 2 | 1 | 16 | 66 | 34 | 12 | 3 | 4 | 2 | 1 | 4 | 5 | 13 | 40 | 3 | 1 | 1 | 1 | |||
JDAP | 11 | 2 | 21 | 31 | 5 | 22 | 5 | 9 | 15 | 8 | 12 | 1 | 1 | 3 | 19 | 12 | 10 | 7 | 6 | 10 | 1 |
JEI | 27 | 3 | 26 | 43 | 24 | 33 | 7 | 8 | 18 | 9 | 15 | 6 | 7 | 13 | 10 | 3 | 3 | 19 | 9 | 1 | |
JEMI | 8 | 4 | 7 | 2 | 1 | 2 | 5 | 1 | 13 | 3 | 1 | 1 | 9 | ||||||||
JIG | 30 | 17 | 117 | 185 | 37 | 152 | 50 | 48 | 86 | 31 | 94 | 53 | 12 | 51 | 31 | 24 | 27 | 41 | 43 | 64 | 30 |
JRS | 18 | 2 | 19 | 2 | 8 | 1 | 7 | 8 | 12 | 2 | 1 | 2 | 67 | 6 | |||||||
PRAI | 7 | 3 | 8 | 11 | 5 | 51 | 22 | 11 | 40 | 15 | 25 | 12 | 2 | 10 | 1 | 1 | 50 | 5 | 2 | 8 | 2 |
SP | 4 | 8 | 29 | 45 | 14 | 37 | 6 | 5 | 18 | 10 | 10 | 2 | 4 | 21 | 9 | 6 | 9 | 5 | 4 | ||
Summary | 194 | 116 | 325 | 538 | 191 | 477 | 157 | 132 | 259 | 132 | 269 | 131 | 33 | 120 | 157 | 128 | 146 | 163 | 174 | 151 | 39 |
Figure 4. Comparisons ofpublication quantities for image coding and image segmentation in last 10 years
FUTURE TRENDS
The field of IE has changed enormously in recent years.
Many techniques have been developed, exploited, or applied only in the last decade. We now see techniques for IE being implemented and used on a scale few would have predicted a decade ago. It is also likely that these techniques will find many new applications in the future.
Viewing the prospective of IE, the work for survey on IE could also be pushed deeply, at least, in two ways. First is that since this survey provides an up-to-date picture regarding IE and its research advance, so further research could be advanced and promoted in appropriate directions. Second is that according to the principles and methods of bibliometrics, a systematic investigation of the factors of the articles indexed in the survey series could be made. This can include the number of authors, the author productivity, the number of collaborative publications, the average number of authors per paper, the active author group, and the author variation ratio, and so forth. Some preliminary works have been performed (Zhang & Li, 2000b, 2001b); an up-to-date and completed version is in preparation. Such a work would reveal the level, status, and alteration of researchers in IE, as well as provide useful information for summarizing the development, progress, trends, and application areas of IE.
conclusion
This article shows an overview of a survey series on IE made in the last 10 years. The idea behind and consideration on this survey, as well as a thorough summary of obtained statistics are illustrated and discussed. All these provide much of useful information regarding the 10 years’ tendency of fast progresses of IE in China and worldwide.
Such a work not only provides a convenient means for literature searching in IE but also presents a detailed picture of hot research topics in the field. Moreover, it may be useful for developers who want to quickly capture the general trends of development in IE and for potential authors who wish to disseminate widely their research results in the associated communities.
key terms
Image: An entity that was captured by some visual systems in looking at the real world and that can be sensed to produce perception. It is a representation, likeness, or imitation of an object or thing, a vivid or graphic description, something introduced to represent something else.
Image Analysis: One of three layers of image engineering, which is concerned with the extraction of information (by meaningful measurements with descriptive parameters) from an image (especially from interesting objects).
Image Coding: A process for representing an image with some other representations in view of reducing data for storage and/or transmission of this image.
Image Engineering: An integrated discipline/subject comprising the study of all the different branches of image and video techniques.
Image Processing: One of three layers of image engineering, which encompasses processes whose inputs and outputs are both images, with the outputs being improved version of inputs.
Image Segmentation: A process consists of subdividing an image into its constituent parts and extracting these parts of interest (objects) from the image.
Image Understanding: One of three layers of image engineering, which transforms data extracted from images into certain commonly understood descriptions, and makes subsequent decisions and actions according to the interpretation of the images.