1Pamscale User Manual(0) Pamscale User Manual(0)
2
6 pamscale - scale a Netpbm image
7
10 pamscale
11 [
12 scale_factor
13 |
14 {-xyfit | -xyfill | -xysize}
15 cols rows
16 |
17 -reduce reduction_factor
18 |
19 [-xsize=cols | -width=cols | -xscale=factor]
20 [-ysize=rows | -height=rows | -yscale=factor]
21 |
22 -pixels n
23 ]
24 [
25 [-verbose]
26 [
27 -nomix
28 |
29 -filter=functionName [-window=functionName]
30 ]
31 [-linear]
32 }
33 [-reportonly]
34 [pnmfile]
35 Minimum unique abbreviation of option is acceptable. You may use dou‐
38 ble hyphens instead of single hyphen to denote options. You may use
39 white space in place of the equals sign to separate an option name from
40 its value.
41
44 This program is part of Netpbm(1).
45 pamscale scales a Netpbm image by a specified factor, or scales indi‐
47 vidually horizontally and vertically by specified factors.
48 You can either enlarge (scale factor > 1) or reduce (scale factor < 1).
50 pamscale work on multi-image streams, scaling each one independently.
52 But before Netpbm 10.49 (December 2009), it scales only the first image
53 and ignores the rest of the stream.
54 The Scale Factors
57 The options -width, -height, -xsize, -ysize, -xscale, -yscale, -xyfit,
58 -xyfill, -reduce, and -pixels control the amount of scaling. For back‐
59 ward compatibility, there is also -xysize and the scale_factor argu‐
60 ment, but you shouldn't use those.
61 -width and -height specify the width and height in pixels you want the
63 resulting image to be. See below for rules when you specify one and
64 not the other.
65 -xsize and -ysize are synonyms for -width and -height, respectively.
67 -xscale and -yscale tell the factor by which you want the width and
69 height of the image to change from source to result (e.g. -xscale 2
70 means you want to double the width; -xscale .5 means you want to halve
71 it). See below for rules when you specify one and not the other.
72 When you specify an absolute size or scale factor for both dimensions,
74 pamscale scales each dimension independently without consideration of
75 the aspect ratio.
76 If you specify one dimension as a pixel size and don't specify the
78 other dimension, pamscale scales the unspecified dimension to preserve
79 the aspect ratio.
80 If you specify one dimension as a scale factor and don't specify the
82 other dimension, pamscale leaves the unspecified dimension unchanged
83 from the input.
84 If you specify the scale_factor parameter instead of dimension options,
86 that is the scale factor for both dimensions. It is equivalent to
87 -xscale=scale_factor -yscale=scale_factor.
88 Specifying the -reduce reduction_factor option is equivalent to speci‐
90 fying the scale_factor parameter, where scale_factor is the reciprocal
91 of reduction_factor.
92 -xyfit specifies a bounding box. pamscale scales the input image to
94 the largest size that fits within the box, while preserving its aspect
95 ratio. -xysize is a synonym for this. Before Netpbm 10.20 (January
96 2004), -xyfit did not exist, but -xysize did.
97 -xyfill is similar, but pamscale scales the input image to the smallest
99 size that completely fills the box, while preserving its aspect ratio.
100 This option has existed since Netpbm 10.20 (January 2004).
101 -pixels specifies a maximum total number of output pixels. pamscale
103 scales the image down to that number of pixels. If the input image is
104 already no more than that many pixels, pamscale just copies it as out‐
105 put; pamscale does not scale up with -pixels.
106 If you enlarge by a factor of 3 or more, you should probably add a
108 pnmsmooth step; otherwise, you can see the original pixels in the
109 resulting image.
110 -reportonly
112 The option -reportonly causes pamscale not to scale the image, but
114 instead to report to Standard Output what scaling the options and the
115 input image dimensions indicate. For example, if you specify
116 -xyfill 100 100 -reportonly
117 and the input image is 500 x 400, pamscale tells you that this means
118 scaling by .25 to end up with a 125 x 100 image.
119 The output is intended to be convenient for machine processing. In the
121 example above, it would be
122 500 400 0.250000 0.250000 125 100
124 The output is a single line of text per input image, with blank-sepa‐
126 rated tokens as follows.
127 · input width in pixels, decimal unsigned integer
131 · input height in pixels, decimal unsigned integer
133 · horizontal scale factor, floating point decimal, unsigned
135 · vertical scale factor, floating point decimal, unsigned
137 · output width in pixels, decimal unsigned integer
139 · output height in pixels, decimal unsigned integer
141 -reportonly was new in Netpbm 10.86 (March 2019).
144 Usage Notes
148 A useful application of pamscale is to blur an image. Scale it down
149 (without -nomix) to discard some information, then scale it back up
150 using pamstretch.
151 Or scale it back up with pamscale and create a "pixelized" image, which
153 is sort of a computer-age version of blurring.
154 Transparency
158 pamscale understands transparency and properly mixes pixels considering
159 the pixels' transparency.
160 Proper mixing does not mean just mixing the transparency value and the
162 color component values separately. In a PAM image, a pixel which is
163 not opaque represents a color that contains light of the foreground
164 color indicated explicitly in the PAM and light of a background color
165 to be named later. But the numerical scale of a color component sample
166 in a PAM is as if the pixel is opaque. So a pixel that is supposed to
167 contain half-strength red light for the foreground plus some light from
168 the background has a red color sample that says full red and a trans‐
169 parency sample that says 50% opaque. In order to mix pixels, you have
170 to first convert the color sample values to numbers that represent
171 amount of light directly (i.e. multiply by the opaqueness) and after
172 mixing, convert back (divide by the opaqueness).
173 Input And Output Image Types
176 pamscale produces output of the same type (and tuple type if the type
177 is PAM) as the input, except if the input is PBM. In that case, the
178 output is PGM with maxval 255. The purpose of this is to allow mean‐
179 ingful pixel mixing. Note that there is no equivalent exception when
180 the input is PAM. If the PAM input tuple type is BLACKANDWHITE, the
181 PAM output tuple type is also BLACKANDWHITE, and you get no meaningful
182 pixel mixing.
183 If you want PBM output with PBM input, use pamditherbw to convert pam‐
185 scale's output to PBM. Also consider pbmreduce.
186 pamscale's function is essentially undefined for PAM input images that
188 are not of tuple type RGB, GRAYSCALE, BLACKANDWHITE, or the _ALPHA
189 variations of those. (By standard Netpbm backward compatibility, this
190 includes PBM, PGM, and PPM images).
191 You might think it would have an obvious effect on other tuple types,
193 but remember that the aforementioned tuple types have gamma-adjusted
194 sample values, and pamscale uses that fact in its calculations. And it
195 treats a transparency plane different from any other plane.
196 pamscale does not simply reject unrecognized tuple types because
198 there's a possibility that just by coincidence you can get useful func‐
199 tion out of it with some other tuple type and the right combination of
200 options (consider -linear in particular).
201 Methods Of Scaling
205 There are numerous ways to scale an image. pamscale implements a bunch
206 of them; you select among them with invocation options.
207 Pixel Mixing
209 Pamscale's default method is pixel mixing. To understand this, imagine
211 the source image as composed of square tiles. Each tile is a pixel and
212 has uniform color. The tiles are all the same size. Now take a trans‐
213 parent sheet the size of the target image, marked with a square grid of
214 tiles the same size. Stretch or compress the source image to the size
215 of the sheet and lay the sheet over the source.
216 Each cell in the overlay grid stands for a pixel of the target image.
218 For example, if you are scaling a 100x200 image up by 1.5, the source
219 image is 100 x 200 tiles, and the transparent sheet is marked off in
220 150 x 300 cells.
221 Each cell covers parts of multiple tiles. To make the target image,
223 just color in each cell with the color which is the average of the col‐
224 ors the cell covers -- weighted by the amount of that color it covers.
225 A cell in our example might cover 4/9 of a blue tile, 2/9 of a red
226 tile, 2/9 of a green tile, and 1/9 of a white tile. So the target
227 pixel would be somewhat unsaturated blue.
228 When you are scaling up or down by an integer, the results are simple.
230 When scaling up, pixels get duplicated. When scaling down, pixels get
231 thrown away. In either case, the colors in the target image are a sub‐
232 set of those in the source image.
233 When the scale factor is weirder than that, the target image can have
235 colors that didn't exist in the original. For example, a red pixel
236 next to a white pixel in the source might become a red pixel, a pink
237 pixel, and a white pixel in the target.
238 This method tends to replicate what the human eye does as it moves
240 closer to or further away from an image. It also tends to replicate
241 what the human eye sees, when far enough away to make the pixelization
242 disappear, if an image is not made of pixels and simply stretches or
243 shrinks.
244 Discrete Sampling
246 Discrete sampling is basically the same thing as pixel mixing except
248 that, in the model described above, instead of averaging the colors of
249 the tiles the cell covers, you pick the one color that covers the most
250 area.
251 The result you see is that when you enlarge an image, pixels get dupli‐
253 cated and when you reduce an image, some pixels get discarded.
254 The advantage of this is that you end up with an image made from the
256 same color palette as the original. Sometimes that's important.
257 The disadvantage is that it distorts the picture. If you scale up by
259 1.5 horizontally, for example, the even numbered input pixels are dou‐
260 bled in the output and the odd numbered ones are copied singly. If you
261 have a bunch of one pixel wide lines in the source, you may find that
262 some of them stretch to 2 pixels, others remain 1 pixel when you
263 enlarge. When you reduce, you may find that some of the lines disap‐
264 pear completely.
265 You select discrete sampling with pamscale's -nomix option.
267 Actually, -nomix doesn't do exactly what I described above. It does
269 the scaling in two passes - first horizontal, then vertical. This can
270 produce slightly different results.
271 There is one common case in which one often finds it burdensome to have
273 pamscale make up colors that weren't there originally: Where one is
274 working with an image format such as GIF that has a limited number of
275 possible colors per image. If you take a GIF with 256 colors, convert
276 it to PPM, scale by .625, and convert back to GIF, you will probably
277 find that the reduced image has way more than 256 colors, and therefore
278 cannot be converted to GIF. One way to solve this problem is to do the
279 reduction with discrete sampling instead of pixel mixing. Probably a
280 better way is to do the pixel mixing, but then color quantize the
281 result with pnmquant before converting to GIF.
282 When the scale factor is an integer (which means you're scaling up),
284 discrete sampling and pixel mixing are identical -- output pixels are
285 always just N copies of the input pixels. In this case, though, con‐
286 sider using pamstretch instead of pamscale to get the added pixels
287 interpolated instead of just copied and thereby get a smoother enlarge‐
288 ment.
289 pamscale's discrete sampling is faster than pixel mixing, but pamen‐
291 large is faster still. pamenlarge works only on integer enlargements.
292 discrete sampling (-nomix) was new in Netpbm 9.24 (January 2002).
294 Resampling
297 Resampling assumes that the source image is a discrete sampling of some
299 original continuous image. That is, it assumes there is some non-pix‐
300 elized original image and each pixel of the source image is simply the
301 color of that image at a particular point. Those points, naturally,
302 are the intersections of a square grid.
303 The idea of resampling is just to compute that original image, then
305 sample it at a different frequency (a grid of a different scale).
306 The problem, of course, is that sampling necessarily throws away the
308 information you need to rebuild the original image. So we have to make
309 a bunch of assumptions about the makeup of the original image.
310 You tell pamscale to use the resampling method by specifying the -fil‐
312 ter option. The value of this option is the name of a function, from
313 the set listed below.
314 To explain resampling, we are going to talk about a simple one dimen‐
316 sional scaling -- scaling a single row of grayscale pixels horizon‐
317 tally. If you can understand that, you can easily understand how to do
318 a whole image: Scale each of the rows of the image, then scale each of
319 the resulting columns. And scale each of the color component planes
320 separately.
321 As a first step in resampling, pamscale converts the source image,
323 which is a set of discrete pixel values, into a continuous step func‐
324 tion. A step function is a function whose graph is a staircase-y
325 thing.
326 Now, we convolve the step function with a proper scaling of the filter
328 function that you identified with -filter. If you don't know what the
329 mathematical concept of convolution (convolving) is, you are officially
330 lost. You cannot understand this explanation. The result of this con‐
331 volution is the imaginary original continuous image we've been talking
332 about.
333 Finally, we make target pixels by picking values from that function.
335 To understand what is going on, we use Fourier analysis:
337 The idea is that the only difference between our step function and the
339 original continuous function (remember that we constructed the step
340 function from the source image, which is itself a sampling of the orig‐
341 inal continuous function) is that the step function has a bunch of high
342 frequency Fourier components added. If we could chop out all the
343 higher frequency components of the step function, and know that they're
344 all higher than any frequency in the original function, we'd have the
345 original function back.
346 The resampling method assumes that the original function was sampled at
348 a high enough frequency to form a perfect sampling. A perfect sampling
349 is one from which you can recover exactly the original continuous func‐
350 tion. The Nyquist theorem says that as long as your sample rate is at
351 least twice the highest frequency in your original function, the sam‐
352 pling is perfect. So we assume that the image is a sampling of some‐
353 thing whose highest frequency is half the sample rate (pixel resolu‐
354 tion) or less. Given that, our filtering does in fact recover the
355 original continuous image from the samples (pixels).
356 To chop out all the components above a certain frequency, we just mul‐
358 tiply the Fourier transform of the step function by a rectangle func‐
359 tion.
360 We could find the Fourier transform of the step function, multiply it
362 by a rectangle function, and then Fourier transform the result back,
363 but there's an easier way. Mathematicians tell us that multiplying in
364 the frequency domain is equivalent to convolving in the time domain.
365 That means multiplying the Fourier transform of F by a rectangle func‐
366 tion R is the same as convolving F with the Fourier transform of R.
367 It's a lot better to take the Fourier transform of R, and build it into
368 pamscale than to have pamscale take the Fourier transform of the input
369 image dynamically.
370 That leaves only one question: What is the Fourier transform of a rec‐
372 tangle function? Answer: sinc. Recall from math that sinc is defined
373 as sinc(x) = sin(PI*x)/PI*x.
374 Hence, when you specify -filter=sinc, you are effectively passing the
376 step function of the source image through a low pass frequency filter
377 and recovering a good approximation of the original continuous image.
378 Refiltering
380 There's another twist: If you simply sample the reconstructed original
382 continuous image at the new sample rate, and that new sample rate isn't
383 at least twice the highest frequency in the original continuous image,
384 you won't get a perfect sampling. In fact, you'll get something with
385 ugly aliasing in it. Note that this can't be a problem when you're
386 scaling up (increasing the sample rate), because the fact that the old
387 sample rate was above the Nyquist level means so is the new one. But
388 when scaling down, it's a problem. Obviously, you have to give up
389 image quality when scaling down, but aliasing is not the best way to do
390 it. It's better just to remove high frequency components from the
391 original continuous image before sampling, and then get a perfect sam‐
392 pling of that.
393 Therefore, pamscale filters out frequencies above half the new sample
395 rate before picking the new samples.
396 Approximations
398 Unfortunately, pamscale doesn't do the convolution precisely. Instead
400 of evaluating the filter function at every point, it samples it --
401 assumes that it doesn't change any more often than the step function
402 does. pamscale could actually do the true integration fairly easily.
403 Since the filter functions are built into the program, the integrals of
404 them could be too. Maybe someday it will.
405 There is one more complication with the Fourier analysis. sinc has
407 nonzero values on out to infinity and minus infinity. That makes it
408 hard to compute a convolution with it. So instead, there are filter
409 functions that approximate sinc but are nonzero only within a manage‐
410 able range. To get those, you multiply the sinc function by a window
411 function, which you select with the -window option. The same holds for
412 other filter functions that go on forever like sinc. By default, for a
413 filter that needs a window function, the window function is the Black‐
414 man function. Hanning, Hamming, and Kaiser are alternatives.
415 Filter Functions Besides Sinc
417 The math described above works only with sinc as the filter function.
419 pamscale offers many other filter functions, though. Some of these
420 approximate sinc and are faster to compute. For most of them, I have
421 no idea of the mathematical explanation for them, but people do find
422 they give pleasing results. They may not be based on resampling at
423 all, but just exploit the convolution that is coincidentally part of a
424 resampling calculation.
425 For some filter functions, you can tell just by looking at the convolu‐
427 tion how they vary the resampling process from the perfect one based on
428 sinc:
429 The impulse filter assumes that the original continuous image is in
431 fact a step function -- the very one we computed as the first step in
432 the resampling. This is mathematically equivalent to the discrete sam‐
433 pling method.
434 The box (rectangle) filter assumes the original image is a piecewise
436 linear function. Its graph just looks like straight lines connecting
437 the pixel values. This is mathematically equivalent to the pixel mix‐
438 ing method (but mixing brightness, not light intensity, so like pam‐
439 scale -linear) when scaling down, and interpolation (ala pamstretch)
440 when scaling up.
441 Gamma
443 pamscale assumes the underlying continuous function is a function of
445 brightness (as opposed to light intensity), and therefore does all this
446 math using the gamma-adjusted numbers found in a PNM or PAM image. The
447 -linear option is not available with resampling (it causes pamscale to
448 fail), because it wouldn't be useful enough to justify the implementa‐
449 tion effort.
450 Resampling (-filter) was new in Netpbm 10.20 (January 2004).
452 The filter and window functions
454 Here is a list of the function names you can specify for the -filter or
456 -windowoption. For most of them, you're on your own to figure out just
457 what the function is and what kind of scaling it does. These are com‐
458 mon functions from mathematics. Note that some of these make sense
459 only as filter functions and some make sense only as window funcions.
460 point The graph of this is a single point at X=0, Y=1.
464 box The graph of this is a rectangle sitting on the X axis and cen‐
467 tered on the Y axis with height 1 and base 1.
468 triangle
471 The graph of this is an isosceles triangle sitting on the X axis
472 and centered on the Y axis with height 1 and base 2.
473 quadratic
476 cubic
478 catrom
480 mitchell
482 gauss
484 sinc
486 bessel
488 hanning
490 hamming
492 blackman
494 kaiser
496 normal
498 hermite
500 lanczos
502 Not documented
503 Linear vs Gamma-adjusted
508 The pixel mixing scaling method described above involves intensities of
509 pixels (more precisely, it involves individual intensities of primary
510 color components of pixels). But the PNM and PNM-equivalent PAM image
511 formats represent intensities with gamma-adjusted numbers that are not
512 linearly proportional to intensity. So pamscale, by default, performs
513 a calculation on each sample read from its input and each sample writ‐
514 ten to its output to convert between these gamma-adjusted numbers and
515 internal intensity-proportional numbers.
516 Sometimes you are not working with true PNM or PAM images, but rather a
518 variation in which the sample values are in fact directly proportional
519 to intensity. If so, use the -linear option to tell pamscale this.
520 pamscale then will skip the conversions.
521 The conversion takes time. In one experiment, it increased by a factor
523 of 10 the time required to reduce an image. And the difference between
524 intensity-proportional values and gamma-adjusted values may be small
525 enough that you would barely see a difference in the result if you just
526 pretended that the gamma-adjusted values were in fact intensity-propor‐
527 tional. So just to save time, at the expense of some image quality,
528 you can specify -linear even when you have true PPM input and expect
529 true PPM output.
530 For the first 13 years of Netpbm's life, until Netpbm 10.20 (January
532 2004), pamscale's predecessor pnmscale always treated the PPM samples
533 as intensity-proportional even though they were not, and drew few com‐
534 plaints. So using -linear as a lie is a reasonable thing to do if
535 speed is important to you. But if speed is important, you also should
536 consider the -nomix option and pnmscalefixed.
537 Another technique to consider is to convert your PNM image to the lin‐
539 ear variation with pnmgamma, run pamscale on it and other transforma‐
540 tions that like linear PNM, and then convert it back to true PNM with
541 pnmgamma -ungamma. pnmgamma is often faster than pamscale in doing the
542 conversion.
543 With -nomix, -linear has no effect. That's because pamscale does not
545 concern itself with the meaning of the sample values in this method;
546 pamscale just copies numbers from its input to its output.
547 Precision
551 pamscale uses floating point arithmetic internally. There is a speed
552 cost associated with this. For some images, you can get the acceptable
553 results (in fact, sometimes identical results) faster with pnmscale‐
554 fixed, which uses fixed point arithmetic. pnmscalefixed may, however,
555 distort your image a little. See the pnmscalefixed user manual for a
556 complete discussion of the difference.
557
560 pnmscalefixed(1), pamstretch(1), pamditherbw(1), pbmreduce(1), pbmp‐
561 scale(1), pamenlarge(1), pnmsmooth(1), pamcut(1), pnmgamma(1), pnm‐
562 scale(1), pnm(1), pam(1)
563
566 pamscale was new in Netpbm 10.20 (January 2004). It was adapted from,
567 and obsoleted, pnmscale. pamscale's primary difference from pnmscale
568 is that it handles the PAM format and uses the "pam" facilities of the
569 Netpbm programming library. But it also added the resampling class of
570 scaling method. Furthermore, it properly does its pixel mixing arith‐
571 metic (by default) using intensity-proportional values instead of the
572 gamma-adjusted values the pnmscale uses. To get the old pnmscale
573 arithmetic, you can specify the -linear option.
574 The intensity proportional stuff came out of suggestions by Adam M
576 Costello in January 2004.
577 The resampling algorithms are mostly taken from code contributed by
579 Michael Reinelt in December 2003.
580 The version of pnmscale from which pamscale was derived, itself evolved
582 out of the original Pbmplus version of pnmscale by Jef Poskanzer (1989,
583 1991). But none of that original code remains.
584
586 This manual page was generated by the Netpbm tool 'makeman' from HTML
587 source. The master documentation is at
588 http://netpbm.sourceforge.net/doc/pamscale.html
590netpbm documentation 03 January 2019 Pamscale User Manual(0)