| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Perspective |
1 Department of Radiology, University of Washington School of Medicine,
Harborview Medical Center, 325 Ninth Ave., Box 359728, Seattle, WA
98104-2499.
2 Department of Radiology, University of Washington School of Medicine,
University of Washington Roosevelt Radiology, 4245 Roosevelt Way N.E., Box
354755, Seattle, WA 98105.
Received October 9, 2002;
accepted after revision November 13, 2002.
Address correspondence to E. J. Stern.
Introduction
 Top Introduction DefinitionsSoftware and Hardware...Image SourceSteps in Image ProcessingAutomationKeep a BackupTreat Each Scan IndividuallyPractice Makes PerfectReferences |
|---|
|
TopIntroductionDefinitionsSoftware and Hardware...Image SourceSteps in Image ProcessingAutomationKeep a BackupTreat Each Scan IndividuallyPractice Makes PerfectReferences |
|---|
Image Size
Image size is a slippery and often misused term. A "large"
image could refer to an image with a large pixel dimension (e.g., 2000 x
3000), one with a large file size (requiring much storage space on a hard
disk), or a large physical size (e.g., 3 x 4 feet). Because of this
potential for confusion, we greatly prefer the terms "pixel
dimensions," "file size," or "physical size"
over the easily misused term "image size." The image file is
determined by pixel dimensions, bit depth, and level of file compression. To
determine the file size of an uncompressed digital image, use the following
formula:
file size = (pixel width x pixel height) x (bit depth / 8).
The result will be the file size in bytes. Divide this by 1024 to determine the size in kilobytes (KB) (and by 1024 again for the size in megabytes [MB]). For example, a 24-bit RGB image that is 459 pixels wide and 612 pixels tall would have a file size of 823 KB:
(459 x 612) x (24 / 8) = 842,724 bytes / 1024 = 823 KB.
An 8-bit gray-scale image that is 459 pixels wide and 612 pixels tall would have a file size of 274 KB:
(459 x 612) x (8 / 8) = 280,908 bytes / 1024 = 274 KB.
Output Resolution
Output resolution is a measurement of clarity or detail of the displayed
image and is expressed as the number of pixels displayed per unit length. This
ratio varies widely, depending on the output device used. On a computer
monitor, the ratio is usually expressed in terms of dots per inch (dpi) or
pixels per inch (ppi). The output resolutions of typical computer monitors
range from 72 to 100 dpi.
Consider an image with pixel dimensions of 1200 x 1800 that is to be output to both a computer screen with an output resolution of 100 dpi and a laser printer with an output resolution of 600 dpi. On the computer screen, the physical size of the image will be 12 x 18 inches. On the laser printer, the physical size of the image will be 2 x 3 inches. Both images have the same pixel dimensions, and hence the same information content.
For another example, imagine that a new computer monitor with a 20-inch-wide screen is needed. Two monitors are contemplated, differing only in output resolution: one will display 1200 x 800 pixels and the other will display 1800 x 1200 pixels. The first monitor will have an image resolution of 1200 divided by 20 or 60 dpi, and the second monitor will have a resolution of 1800 divided by 20 or 90 dpi.
Software and Hardware Requirements
|
TopIntroductionDefinitionsSoftware and Hardware...Image SourceSteps in Image ProcessingAutomationKeep a BackupTreat Each Scan IndividuallyPractice Makes PerfectReferences |
|---|
Either Macintosh (Apple, Cupertino, CA), Windows (Microsoft, Redmond, WA), or UNIX-based computers can perform the image manipulation. The interfaces might differ slightly among platforms, but the concepts are the same. Image processing can push a system's limits like few other tasks. Therefore, a computer with the fastest processor speed affordable (500 Mhz minimum) should be purchased. Macintosh and Windows are the most popular platforms, but the actual platform chosen is a largely a matter of preference. To optimize a system for image processing, one should consider upgrading the system RAM (random access memory), the display RAM (video RAM [VRAM]), and the hard drive. An amount of RAM that is 3–5 times the size of the image file being used, plus 5–10 MB more, should be allocated to Photoshop. With large images, this memory requirement can add up quickly. Fortunately, RAM is currently inexpensive, and a 256-MB RAM upgrade card can be purchased for $65 in United States currency. Besides system RAM, at least 6–8 MB of VRAM, the specialized memory chips used to hold the image for computer display, is needed. Because Photoshop also uses the hard disk intensively as scratch memory, a large fast disk drive is recommended.
|
TopIntroductionDefinitionsSoftware and Hardware...Image SourceSteps in Image ProcessingAutomationKeep a BackupTreat Each Scan IndividuallyPractice Makes PerfectReferences |
|---|
An increasingly popular source of digital images is the digital camera. At this writing, one of the authors uses a Sony DSC-50 (2.1-megapixel) camera (Sony of America, New York, NY) and the other uses an Olympus C-2500L (2.5-megapixel) camera (Olympus America, Melville, NY). Images can also be downloaded directly from the Internet. Graphics downloaded or saved from Web pages may not be acceptable for print products because these graphics have relatively low pixel dimensions, which are satisfactory for screen display but are often too low for printed output.
|
TopIntroductionDefinitionsSoftware and Hardware...Image SourceSteps in Image ProcessingAutomationKeep a BackupTreat Each Scan IndividuallyPractice Makes PerfectReferences |
|---|
Step 1
Convert to gray scale. Most radiologic images should be converted to gray
scale unless they are true color images. This conversion immediately cuts the
file size by two thirds without any loss of image quality and is performed by
opening the "Image" pop-up menu, selecting "Mode," and
then selecting "Gray Scale."
Step 2
Crop your image as necessary and remove unwanted or distracting data such
as patient identifiers, nonpertinent body parts, white borders, or artifacts.
This step will also help reduce file size. Patient identifiers can be covered
with a black box, or the text can be cut out after a black background has been
selected.
Step 3
Adjust the "Levels" to take advantage of the entire gray scale
available. This is one of the most important steps. You can use the
"Auto Levels" function from the "Image" pop-up menu,
but we recommend fine-tuning your image with the "Levels"
adjustment (Figs. 1A,
1B,
1C).
|
|
|
Step 4
Rescale your image to the appropriate physical size for the
"Output" device you have chosen. Use the "Image"
pop-up menu to select the image size and the new dialog box to adjust the
pixel dimensions and print size as needed. For example, consider a 512 x
512 pixel CT image destined for both a Web site and a journal article. If the
image was not rescaled, it would appear to be 6.83 inches wide (512/75) on a
75-dpi monitor screen but only 1.70 inches wide (512/300) when printed at 300
dpi in a journal. Even though both images contain exactly the same number of
pixels (Figs. 2A,
2B,
2C,
3A,
3B,
3C,
3D,
3E,
3F,
3G,
4A,
4B,
4C,
4D), for the Web site, this
output size may be too large; for the journal article, the output size would
be too small. Rescaling works by throwing away pixels or adding new pixels to
the image file. If every other pixel is thrown away, the image appears one
half as large on the output device. If one new row and column of pixels in
between each of the original pixel rows and columns are added, the image will
appear twice as large on the output device. These pixels are subtracted and
added using a mathematic process called interpolation, designed to add or
remove pixels to blend them smoothly with the original pixels.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
The general minimal requirements for output resolution for paper publication are 1200 dpi/ppi for monochrome images. This resolution applies to images that are purely black and white, such as line graphs. For halftones, 300 dpi is acceptable (CMYK and gray scale). This output resolution is for images containing pictures only, not text-labeling or thin lines. For combination halftones, 600 dpi/ppi is appropriate. This resolution is for images containing pictures and text-labeling or thin lines or both.
For print, some publishers may require saving images in exact layout size so that scaling is not necessary. This requirement varies by publisher and journal, but the publisher may require that the image be submitted in picas (6 picas/inch). A common size might be 19.6-pica width for one column and 30- to 41-pica width for two-column images. Setting picas in Photoshop is easily performed in the same pop-up menu as that used for adjusting size in inches.
Step 5
Save the image in an appropriate file format. Images can be saved in a
variety of file types, including TIFF (tagged image file format), which is
preferred for print because there is no loss of image information from
compression: JPEG (Joint Photographic Experts Group); GIF (graphic interchange
format); or PNG (portable network graphics), which are preferred for Web
publication (Table 1). File
compression can be used to reduce the file size for storage or for
transmission across the Web. This is performed using compression software that
removes redundant information from the file, which is added back to the file
when it is decompressed for viewing. An image decompressed and viewed after
lossless compression will be identical to its state before being compressed,
such as with TIFF. An image decompressed and viewed after lossy compression
will be similar but not completely identical to the source image, such as with
JPEG. The degree of compression affects the image-quality degradation.
Repeated lossy compression causes additive image degradation. Figures
5A,
5B,
5C and
6A,
6B illustrate the relative
effects of lossless and lossy compression on the quality of digital
images.
|
|
|
|
|
|
TIFF files can include color or gray-scale information. GIF images are generated specifically for display on a computer monitor. JPEG is a lossy compression format because it deletes redundant information from an image. JPEG compression can range from small to large amounts of lossy compression and is commonly used on the Web. Lossless compression of images typically reduces the file size by approximately 50–70%. Significantly greater storage savings can be achieved by compromising some quality using lossy compression. In practice, 1:7-JPEG compression can be used with little or no perceptible loss in image quality, even for a difficult to reproduce chest radiograph (Figs. 6A, 6B). The amount of JPEG compression is selected when performing a "Save As" from the "File" pop-up menu, selecting the "JPEG" format, and then selecting the relative amount of compression requested in the dialog box.
If images are being used only for electronic slide presentations, skip step 6 and proceed to step 7.
Step 6
The halftone process.—Digital image manipulation for general
slide and Web consumption is different from that considered press ready (i.e.,
ink on paper). Gray ink does not exist—all shades of gray between black
and white must be rendered with black ink. This process is called halftoning
and results in a moderate decrease in the output resolution of the final
image. Because of this phenomenon, printers require a source image that
contains 40–100% more information (pixel dimensions) than appears on the
final image. This halftone factor may vary from printer to printer. The
halftone conversion process is performed by the printer, so the main
responsibility of the author is to supply an image with sufficiently detailed
pixel dimensions, usually by sending the images at 300 ppi.
Dot gain.—For images that will be published in print, one must also take into account the dot gain that occurs when putting ink on paper [1]. When ink is transferred to paper during the printing process, it tends to spread though the paper for a small distance and make the final dot a bit larger. This increase in dot size is called "dot gain" and depends on the porosity of the paper. For example, porous paper has more dot gain, whereas with glossy paper, less ink is absorbed, thus less dot gain. The type of paper varies, depending on the purpose. One type is newsprint, a coarse paper made mostly from wood pulp that is highly porous and manufactured almost exclusively for printing newspapers. The dot gain for newsprint is 20% or more. Uncoated paper is unvarnished. Dot gain for uncoated paper is roughly 12%. Coated paper has a varnish coat that helps to seal the paper and reduce dot gain (which runs at about 8% on average). High-quality gray-scale and four-color images are printed on coated stock. Calendering is a normal process for finishing most paper. Supercalendering is performed for some papers to obtain a smoother less porous surface. Supercalendered paper can be coated or uncoated. The AJR uses coated supercalendered paper (Sterling Ultra Web Gloss 60#, MeadWestvaco, Dayton, OH).
Failing to take dot gain into account is a common mistake when submitting electronic art-work and, in printed digital images, may cause images of poor quality (Figs. 7A, 7B, 7C, 7D, 7E) that look stark, with obliteration of fine detail. To correct for dot gain, use the "Curves" function, found under the "Image" pop-up menu, "Adjust" submenu. Adjusting the "Input" and "Output" to "95% White" and "95% Black" will account for dot gain in printing most radiology images (Figs. 7A, 7B, 7C, 7D, 7E), although the outcome varies slightly by journal [1], depending on the type of paper used.
|
|
|
|
|
Dot gain is also an important factor in determining screen frequency for halftoning. If the paper is porous, dot gain will detract from any detail that could be gained from using a high screen frequency. Dot gain is controlled by strict quality control measures that ensure that halftone quality is adequate.
Step 7
During many image manipulations, such as the initial digitalization process
itself, rescaling, and the halftoning processes, tiny amounts of image detail
are lost. Some of this image sharpness can be restored by various
image-sharpening filters. The filter used by most graphics professionals has
the unlikely name of "Unsharp Mask." The downside of image
sharpening is that it adds a tiny bit of noise to the image. Choosing
appropriate settings of the "Unsharp Mask" filter can
significantly improve the image sharpness with only minimal increase in
perceptible noise. With 300- ppi images (considered high-resolution), use a
radius setting approximately equal to the output resolution divided by 200.
This usually equates to a radius of 1.2. Use a threshold between 2 and 6
(usually 3).
Unsharp masking should be performed just before the final save. Always apply "Unsharp Mask" to a copy of the image, never to the original image (Figs. 8A, 8B).
|
|
|
TopIntroductionDefinitionsSoftware and Hardware...Image SourceSteps in Image ProcessingAutomationKeep a BackupTreat Each Scan IndividuallyPractice Makes PerfectReferences |
|---|
After the "Action" is recorded, choose the "Automate" "Batch" command from the file pop-up menu to perform the macro on an entire folder of images in just seconds.
|
TopIntroductionDefinitionsSoftware and Hardware...Image SourceSteps in Image ProcessingAutomationKeep a BackupTreat Each Scan IndividuallyPractice Makes PerfectReferences |
|---|
|
TopIntroductionDefinitionsSoftware and Hardware...Image SourceSteps in Image ProcessingAutomationKeep a BackupTreat Each Scan IndividuallyPractice Makes PerfectReferences |
|---|
|
TopIntroductionDefinitionsSoftware and Hardware...Image SourceSteps in Image ProcessingAutomationKeep a BackupTreat Each Scan IndividuallyPractice Makes PerfectReferences |
|---|
Last, read the publishers' copyright transfer agreements for digital images. For example, the American Roentgen Ray Society asks the author to certify that "the scientific content of radiologic images has not been altered, and that disclosure has been made regarding computer enhancement or other electronic manipulation of radiologic images" [3].
|
TopIntroductionDefinitionsSoftware and Hardware...Image SourceSteps in Image ProcessingAutomationKeep a BackupTreat Each Scan IndividuallyPractice Makes PerfectReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
This article has been cited by other articles:
|
|
 |
|
  C.-S. Yam Freeware Program for Converting Image Sequences to Flash Movie Loops Am. J. Roentgenol., January 1, 2008; 190(1): W68 - W72. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
C.-S. Yam, J. Kruskal, and M. Larson Creating Animated GIF Files for Electronic Presentations Using Photoshop Am. J. Roentgenol., May 1, 2007; 188(5): W485 - W490. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
G. Gunabushanam and S. Sharma Simple Method to Eliminate Camera Reflections in Digital Photographs of Single-Emulsion Films Am. J. Roentgenol., January 1, 2007; 188(1): W96 - W97. [Full Text] [PDF]
|
|
|
|
|
|
C.-S. Yam Using macromedia flash for electronic presentations: a new alternative. Am. J. Roentgenol., August 1, 2006; 187(2): W209 - W217. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
C.-S. Yam Removing Hidden Patient Data from Digital Images in PowerPoint Am. J. Roentgenol., December 1, 2005; 185(6): 1659 - 1662. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
C.-S. Yam Using PowerPoint to Create High-Resolution Images for Journal Publications Am. J. Roentgenol., July 1, 2005; 185(1): 273 - 276. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
O. Lababede Digital Image Editing Using PowerPoint: Part 1, Introduction to the Image-Manipulation Commands Am. J. Roentgenol., September 1, 2004; 183(3): 847 - 851. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
O. Lababede Digital Image Editing Using PowerPoint: Part 2, Practical Applications of the Image-Manipulation Commands Am. J. Roentgenol., September 1, 2004; 183(3): 853 - 858. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
