Software for enlarging pictures

I want to enlarge a small photo without losing quality. Is that possible?

On shows like CSI they have programs that do that...not sure if its real.

 

Impossible. Sure you can resize an image to a smaller size without losing quality, but making it bigger than its native resolution and retaining the same quality is impossible.

 

People who work with Real Software absolutely HATE shows like CSI and Law&Order. Total fiction.

 

It really depends on the image itself. If it's really simple, you can convert it to vector image, smooth th edges, and enlarge it to infinity without losing any quality.

 

not possible....but i would love to see it....

 

if you have a photoshop CS series, get a plug in called Genuine Fractals. this can do the magic. i've been using it for long time.
links : http://www.ononesoftware.com

 

Actually it is possible, within limits of course. You can't take a tiny picture and blow it up to poster size, but you can enlarge photos. I've had great results with the application called ImageEnlarger which is a SourceForge project. I have taken a 800x600 image and increased the width to 1280 with amazing results.

You have to play around with the Sharpness, PreSharpen and DeNoise controls a bit. The user interface is a bit odd. But like I said I was amazed by the results.

http://sourceforge.net/projects/imageenlarger/

Gary

 

nothing beats csi

(and +1 to scuderia.. one can scale up an image without losing anything. and with right filtering, making it still look pretty. but one can't ADD anything. only guess. those guesses can be quite fitting, though.. upscalers in tvs and video software do this, to scale up video to hd, making it look quite pretty)

 

Most decent image viewing programs can scale up images to some extent, although results will vary. I find XNView to have a large variety of resampling techniques and can produce some great results. Give it a try. Go to resize, chose a resolution higher that what the image currently is, choose a resampling algorithm, and see the result.

 

actually, i want you to resize an image to a smaller size without losing quality if you half the res (as an example), it loses 3/4th of the pixels. so no matter what you do, it loses 3/4th the information.

increasing the size without any loss is easy. non-filtered (often called pointsample or similar) zoom upwards always has all the picture information. but not MORE.

 

Yes, but the question was not about not losing information, it was about not losing " quality". Whether or not that is possible very much depends on how you define "quality". After that, it depends on the picture. For generic pictures, and using intuitive definitions for "quality", the answer is usually negative. You will lose some quality compared to the original, but not too much if you're lucky.

P.S.: I liked your CSI collage!

P.P.S.: Note that, for videos, the problem is quite different. Since we also have the time axis available, often enlargements can be produced at much better quality than what would be possible for still images. The mathematics of these procedures is quite interesting, by the way.

 

hm right. fine with that.

and yeah, the video thing is great. seen the upscaling of old vhs to full-hd done by intel. i so hope this one day will be available for consumer end hw. it would be absolutely awesome. (and it would even work a bit csi style.. at least the licence plates of non-readable distances could get made read again. they showed it)

 

Does it lose information or data? The two are not necessarily the same.

Gary

 

I pretty much meant to make the image larger and still clear. The CSI collage shows what im looking for pretty much. Thanks for all the replies, im going to try them out.

 

actually, in that case, they are.

if you have a picture of a chessboard, the black and the white squares each exactly one pixel big, and reduce the size to half, you will lose 3/4th of the squares. those squares contained information/data about the image.

it a) drops them or b) blurs them in some form. in a), you lose actual chessboardsquares, in case b), you lose the information about the differences between some of the squares (bilinear filtering will make 4 squares into one, you essentially get a gray square instead of 2 black and 2 white ones, and have lost ALL contrast.

in both cases, you lost 3/4th of the information. different form of information depending on the filtering, but each time, it would drop exactly 3/4th of it.

simple mathematical thing, independent on the actual image. now it could be true, that this information is unimportant. but take f.e. a picture of a face. with all the skin-pores at around the 1-3 pixel size. reducing the picture to half width/half height would not make the face go away, but the whole skin detail would be lost (like going from Bluray to DVD ).


yeah, i've done quite some information theory about graphics. it's quite interesting once you go into 3d rendering, what sort of 2d projections, samplings, filterings result in what losses, what artefacts, etc.

 

In your example the loss of data does indeed lead to a loss of information as well, because you are already at the minimum resolution necessary to express the information. But what if the squares were each 40X40 pixels? Then a substantial loss in data would not necessarily equate to a loss of information at all. And that is the only point I was trying to make. Data does not equal information. There is a definite relationship between the two, without a doubt. It just depends on the required resolution of data needed to express the information.

In your example that threshold of needed data is one pixel at a depth of one bit. But if we change the chessboard to red and black we need to change the bit depth to retain the fidelity of information required.

This is a battle I fight all the time with management folks. They often times think just because they have data they have information. Nothing could be further from the truth. You could have reams of data, but if it is not aggregated, interpreted or presented correctly there is no information conveyed at all.

But we are getting far afield of the original point of this thread. (Although it is a interesting conversation.)

Gary

 

you got a point there.

but look at it from the developer perspective: the image contains x individual data records. each contains information. not all of that might be useful (your poitn there). but the programmer writing the reduction algorithm can't know. for him, all he knows is, he will use 3/4th of the information.

fact is, if the squares would be 2x2 each, then yes, the picture would look identical (not blending anything). but it still wouldn't have all the information in it that it had before. what information got lost, then?

the information that shows that, indeed, in detail, there IS NO FURTHER INFORMATION. that is information, too. it could have had tiny irregularities that got hidden away by reducing it's size. or it might not. and you lose THAT information. (and this is why CSI is so funny.. they get exactly THAT information back out of the lowres picture).

i agree with you that data != information (!= means not equal), and that this is important to consider.

but the point still stands: you reduce the amount of data, and by that, the amount of information you can get out of it. take the skin-pores example from before. you would not lose the information on WHO is on the photo. but you would lose the information on how clean/messed up his skin is.

 

Yes, but my real assertion here is that only the data that is important is information, everything else is just data. And that's the point I have to drive home to many folks in upper management in the consulting we do. They want to measure everything, but we stress the mantra of don't measure what you can't change. That is where the information is, the rest is all meaningless data.

From the perspective of the developer in your scenario, they would have no idea what is information and what is not. It is all data to them. They will not lose 3/4ths of the information, they will lose 3/4ths of the DATA. That might mean zero loss of information or it might mean 100% loss of the information. There is no way to tell. Much like beauty, information is in the eye of the beholder.

In your final example above, again using the idea of information is in the eye of the beholder, who the person is may very well be the ONLY information in the picture to one person. The details of the skin are just data conveying no information at all. To someone else the higher detail of who the person is may not be information at all, and the only information comes from the very skin details that were noise to the other user.

To site an example of this sort of thing, take a vat that is being heated to a particular temperature. And for the sake of example lets say we need to know the temperature of the vat to within two degrees. And that we don't care if the temperature varies by more than six degrees over a period of two hours. There is no point in putting in place an expensive thermocouple that reads to within one one thousandth of a degree every 2 seconds. All that extra precision (data) contains no information.

Thanks again for a stimulating conversation!!!

Gary

 

i disagree with your belief in "that's the only information" f.e. in the photo of the client.

it might be the only relevant information. but everything else is information, too.

you might have reduced the photo, as it's good enough to "see who it is". then you go to a friend who is doctor about some scratching that person on the photo has, something icky. the doctor says, well, show me the pic in detail, i want to see his skin.

and the photo by now lacks that information.

but yeah, esp. for management, it's important on the information that is RELEVANT for the company. to filter out information nonrelevant for making choices.

but it doesn't degrade other information to non-information. see csi, they can use and abuse any tiny information they find

 

Quite right. But you just made my point: what is information to one person is noise to another. The first person could care less about the blemishes on the persons face. That is noise, not signal. The doctor is only interested in the blemishes.

Only relevant data is information, everything else is data. But what we are talking about is scoping of data. In the case of the picture, we don't know the scope ahead of time. Until the need arises for the doctor to see the photo, the scope was strictly "who is in the picture". The moment the doctor becomes involved, the scope changes. Until that scope was changed there was no need for any of the higher resolution data. When the scope did change, suddenly that data became information. So in that context where the scope may change over time, the importance of retaining the data to a higher resolution might be very important.

Back to my example, if we knew we are going to erase the vat temperature data every day, we know its scope won't change. But if we are going to keep a backlog of the data for ten years, we might want to go ahead and take the readings with more than 2 degrees of precision and on a cycle of every few minutes, because at some time in the future the scope of this data might change and it might become information at such resolution. But until that scope does change it is still just data we have retained. It only becomes information when we put the data to use.

Man, I hope we aren't boring the hell out of everyone else with this conversation since we are so far off the original post. (Well not so far off, I guess.) Anyway, cheers!

Gary

 

well, we just have different interpretations for what is information. but both of us are right.

so... reducing the image will reduce the the resolution of it's data, which means there is less space for possible information. it will cut away some data, which could, at one point, be important information, or not. that is in the eye of the beholder, a thing of interpretation and context.

so i, for savety reasons, define information without the context. any datapoint IS information. which parts of it you want to save, and which one you don't, is different for each user. the image reduction will remove ANY information that is below the target resolutions step size. if there's no information there that you care in your context, then you have lost NOTHING to care about.

but the detail-information is still gone, and, when context changes, it might be missed.

for me, information is independent on the interpretation. the interpretation is a filter. and reducing the amount of actual information to only that, what you care about with your filter, that is the big thing f.e. in video and audio compression.

this might be because i, possibly, am a bit autistic. never checked for it, but i sure have signs of it. i care about all the details that might be unimportant. everytime. i actually focus on them. which is why i love fullhd movies. they contain tons of details, for me, all important information to parse and explore. others don't care, and can watch a movie on VHS just as well. it still brings the message, so why should i have the high quality version?

for me, those detail informations make all the difference in the world.



or in short:

for me
data = information

difference between us:

my information = your data
my useful information = your information

but what is useful, that can change. and once compressed, and the data is gone, you can't get that suddenly new useful information out anymore. no chance.

which is why i still stand by my point: you lose information. it might never matter. but you lost it, if it matters one day.

 

We are really in agreement 100%. I just harp on the narrow definition of information versus data, because of its relevance to what I do. I have to be careful to make the distinction between data and information otherwise we find folks gathering all manner of irrelevant and meaningless data in the mistaken belief that they are gathering information.

In an ideal world I would agree that you should define information without context (scope), but in reality we can't. We have to make some trade offs. Otherwise we would have a single 3 minute digitized song take up an entire DVD... big ol' grin. Both of us being audio "gearheads", I think we can agree that such a scope of data would, most likely, be useless, even in the future. So we don't capture that much detail. 96 bit at 144khz is probably sufficient. But we need to be careful about not underestimating that scope either. Your example of the picture of the pimple faced person is a prime example. If we underestimate the scope of the data we may throw away useful information.

Gary

 

indeed. now the problem is, that exactly in my environment, the additional information often gets, later on, useful information. true in programming very much, but true f.e. in music, just as well.

for final storing, one can really reduce to only what's needed (compress with mp4 and good psychoaccoustics, and it sounds great, using very low amount of storage). but if you have some audio sample, that you might have to process later, you need all the information in it that you can. imagine stretching the audio out for some breakdown (like turning off a playing vinyl disc). suddenly, 144khz might not be enough. at the moment the audio gets played at lower than 30% of it's speed, artefacts will popup.

but yeah, a 3 min song taking up a dvd.. that's very... "save" to not ever have problems

i like to store, at one point, the untouched original highest quality possibly audio version of my stuff. in case of further processing requirements, that's important to not start to show up banding errors (both in time and in loudness direction).

it's quite interesting, most people don't understand that, for final storage, you don't need much data, you can compress the heck out of it. but if you want to further process the data, you need the most uncompressed, highest resolution source you can get (but not higher than that).
which is why f.e. photographers always work with the RAW format. even if, in the end, it's a 128x96 pixel image on a web page

but i agree that making that distinction very strong is VERY useful when talking to others. even while, for one self, one knows that's actually a blurry border.

one thing i try to preach in here: you can do what you want, but always be aware of what you preach. it doesn't have to be the same thing. make sure you preach the stuff in a way that people won't abuse it in some form. (see my statements against tweaking, f.e.). your situation is similar. make a clear, simple defintion of the borders, so no one can step over and mess it up. and i know, management guys are GREAT at doing so


and btw: everything is relative

for us, this thread contains useful information
for the op, it might just be useless noise
for the mods, it's just data on the nbr database