i just thought of this! - DPChallenge Forums

Canon EF-S 18-200mm f/3.5-5.6 IS

07/14/2005 10:11:42 PM · #1

wait a second. let's say you had a 3 mp camera. you took a picture. and then magically somehow you could shift your camera to the right one half-pixel in distance. and then you hit the shutter again. then you put both images in photoshop and made them 50% transparent and then layered both of the pictures together. wouldn't you have a barely fuzzy 6 mp image?

CorySmith

07/14/2005 10:12:45 PM · #2

technically yes
this technique is also called stitching

CEJ

Canon EOS-40D

07/14/2005 10:14:18 PM · #3

I thought stitching was side to side, not top to bottom?

Nikon AF Nikkor 50mm f/1.8D

07/14/2005 10:14:20 PM · #4

Originally posted by Fetor:

technically yes
this technique is also called stitching

i thought stiching was when you took two pictures and put them side by side.

deapee

Kodak V803

07/14/2005 10:16:00 PM · #5

Originally posted by art-inept:

Technically, your PSD (or photoshop file) would consist of 6mp -- but only 3mp would be visible. When I say technically, I mean that your image would consist of two 3mp images, one ontop of the other. So to answer your question, no, the result would not be a 6mp image, once saved in JPG format.

07/14/2005 10:16:26 PM · #6

i didn't make this very clear, i mean putting the pictures on top of eachother not adjacent

CorySmith

07/14/2005 10:16:35 PM · #7

it can be in a vertical manner, think of when you are doing vertical pano's you still are stitching the images together

Canon RF 24-70mm f/2.8 L IS

07/14/2005 10:18:27 PM · #8

hold on, i dont really understand what ur saying, deapee. but i'll brb in 20 minutes it's dinner time for me

kyebosh

Canon EOS-20D

Canon EF 80-200mm f/2.8L

07/14/2005 10:18:40 PM · #9

you might get more detail, but you won't get more MP.

kirbic

Canon EOS R5

07/14/2005 10:19:28 PM · #10

Originally posted by Fetor:

it can be in a vertical manner, think of when you are doing vertical pano's you still are stitching the images together

The OP is talking about putting one image *overlaying* the other, not next to or above/below :-)

kirbic

Canon EOS R5

Canon RF 24-70mm f/2.8 L IS

07/14/2005 10:27:47 PM · #11

Originally posted by art-inept:

You have hit on something. In order to gain the most advantage, you need to:
1.) Know exactly how the alignment changes from image to image
2.) Upsample all the images so that they can be precisely aligned to sub-pixel accuracy

In practice, if you take several shots with small random movement between them, then overlay them carefully, even without upsampling, you will see an increase in detail. This shouldn't surprise anyone, we're gathering several times the information present in a single image, and averaging it. An average, properly performed, is a better estimate of reality than a single sample. Remember that none of the color channels are sampled at the full sensor resolution due to the bayer pattern of the sensor.
You also lose some random noise (half of it, for a four-image stack) which also lets fine detail come through.

Pug-H

Pentax K10D

Sigma 28-80mm f/3.5-5.6 Aspherical HF for Pentax

07/14/2005 10:42:31 PM · #12

I thought stitching was when you ran a long distance and got a pain in your side.

CorySmith

07/14/2005 10:43:37 PM · #13

Originally posted by Pug-H:

I thought stitching was when you ran a long distance and got a pain in your side.

thats cramping

aronya1

Nikon D70

Nikon AF-S DX Zoom-Nikkor 18-70mm f/3.5-4.5G IF-ED

07/14/2005 10:50:49 PM · #14

Originally posted by Fetor:

Originally posted by Pug-H:
I thought stitching was when you ran a long distance and got a pain in your side.

thats cramping

No. That's when you sleep out in the woods with a tent.

07/14/2005 10:56:02 PM · #15

i think you will get more megapixels, just the pixels will be half the size of a regular pixel. so if you double the image size the small pixels should become the same size and same amount of a picture taken with a 6 mp camera.

Canon RF 24-70mm f/2.8 L IS

07/14/2005 10:58:37 PM · #16

Originally posted by kirbic:

Originally posted by art-inept:
wait a second. let's say you had a 3 mp camera. you took a picture. and then magically somehow you could shift your camera to the right one half-pixel in distance. and then you hit the shutter again. then you put both images in photoshop and made them 50% transparent and then layered both of the pictures together. wouldn't you have a barely fuzzy 6 mp image?

You have hit on something. In order to gain the most advantage, you need to:
1.) Know exactly how the alignment changes from image to image
2.) Upsample all the images so that they can be precisely aligned to sub-pixel accuracy

In practice, if you take several shots with small random movement between them, then overlay them carefully, even without upsampling, you will see an increase in detail. This shouldn't surprise anyone, we're gathering several times the information present in a single image, and averaging it. An average, properly performed, is a better estimate of reality than a single sample. Remember that none of the color channels are sampled at the full sensor resolution due to the bayer pattern of the sensor.
You also lose some random noise (half of it, for a four-image stack) which also lets fine detail come through.

what do you mean by "unsample" ?

maxj

Canon EOS-300D Rebel

Canon EF 50mm f/1.8 II

07/14/2005 11:01:22 PM · #17

not that long a go, a friend o mine wanted to do a life size cardboard cutout of a someone. I took three shots, then stitched them together with the Canon Photostitch software. the resulting image was of much higher quality than just one image stretched out.

kirbic

Canon EOS R5

07/14/2005 11:03:09 PM · #18

Originally posted by art-inept:

what do you mean by "unsample" ?

UPsample: to change the pixel dimensions of an image to a higher number, usually an integer multiple of the original size.

Example:
Take a Canon 10D file, the original size is 3072x2048. Upsampling by 2x linear resolution means multiplying the sidth and height by 2, so the image size becomes 6144x4096.

07/14/2005 11:05:58 PM · #19

Originally posted by kirbic:

Originally posted by art-inept:
what do you mean by "unsample" ?

UPsample: to change the pixel dimensions of an image to a higher number, usually an integer multiple of the original size.

Example:
Take a Canon 10D file, the original size is 3072x2048. Upsampling by 2x linear resolution means multiplying the sidth and height by 2, so the image size becomes 6144x4096.

oh got it. i also mentioned this above

Canon 12x IS USM (S3 IS built-in Lens) -- 36-432mm (35mm EQ)

07/14/2005 11:07:53 PM · #20

Originally posted by maxj:

isn't stitching where you put shots next to each other? not putting them on top of eachother?

GeneralE

Canon PowerShot S3 IS

07/14/2005 11:30:24 PM · #21

Originally posted by kirbic:

... we're gathering several times the information present in a single image, and averaging it. An average, properly performed, is a better estimate of reality than a single sample. Remember that none of the color channels are sampled at the full sensor resolution due to the bayer pattern of the sensor.
You also lose some random noise (half of it, for a four-image stack) which also lets fine detail come through.

This is how a lot of astrophotography is done, with several images stacked and processed to achieve a better result than any single capture. A free program called Registax can help you with this.

Canon 12x IS USM (S3 IS built-in Lens) -- 36-432mm (35mm EQ)

07/14/2005 11:41:07 PM · #22

Originally posted by GeneralE:

Originally posted by kirbic:
... we're gathering several times the information present in a single image, and averaging it. An average, properly performed, is a better estimate of reality than a single sample. Remember that none of the color channels are sampled at the full sensor resolution due to the bayer pattern of the sensor.
You also lose some random noise (half of it, for a four-image stack) which also lets fine detail come through.

This is how a lot of astrophotography is done, with several images stacked and processed to achieve a better result than any single capture. A free program called Registax can help you with this.

ya but i thought this wasn't to increase detail but to reveal dimmer stars and whatnot

GeneralE

Canon PowerShot S3 IS

07/14/2005 11:42:21 PM · #23

Originally posted by art-inept:

... ya but i thought this wasn't to increase detail but to reveal dimmer stars and whatnot

It amounts to the same thing.

SDW

Canon EOS-60D

Tamron AF 28-300mm f/3.5-6.3 XR Di LD Aspherical IF for Canon

07/15/2005 12:27:37 AM · #24

You would still have a 3mp photograph. Say you take two frames of the same image using a tripod. The first picture you take at -1/3EV, the second at 0EV you have two 3mp pictures but if you overlay them you still only have a 3mp picture. But you do have more detail to work with. The reason you still have the same amount of mp's is because the image is still the same size (WxH).