4 ms·

I'm not sure the reviewer has a good understanding of how to interpret the Siemens star chart (referred to as a 600 dpi test image). I'm not entirely qualified

I'm not sure the reviewer has a good understanding of how to interpret the Siemens star chart (referred to as a 600 dpi test image). I'm not entirely qualified to interpret the results either, but I know that differing anti-aliasing and image processing routines can produce significant artifacting in this type of image. This must be factored in to any evaluation. My point is that the better perceived image quality of the star chart on the iPhone isn't necessarily an indicator that the "screen quality" is better. It's just as likely that the image processing in the iPhone has been tweaked to produce an image more in line with what we'd expect to see, at the price of accurate reproduction. This type of trade off is made all the time in image processors found in cameras, televisions, and computer displays. Your preference for one or the other has to do with your priorities. Do you value rigid accuracy in reproduction, even if it results in some artifacts, or do you prefer an image that is free of artifacts at the expense of accuracy?

sebbi 15y ago
Also the images don't appear to be fullscreen on any of the phones. If some highres image is being scaled and filtered to fit on the screen, than this "test" is definetly not an indicator of screen quality.
theatrus2 15y ago
The color test was also bunk - no deltaE tracking, no gamut comparisons, no brightness level comparisons.
yardie 15y ago
what we'd expect to see, at the price of accurate reproduction Can you explain what this means. Since our eyes are the only reference we have for interpreting colors (the other being wavelength, and that's just a number) I'm curious what an accurate color really means. I know that we can see billions of colors but only a thousands at the same time.
- bradleyland 15y ago
  I'm referring to the star chart, not the colors. You'll notice the curvy artifacts in the chart that looks like the ones found at this link: http://www.google.com/search?q=siemens+star+chart http://www.google.com/search?q=siemens+star+chart Those curves are an illusion. They don't actually appear in the image. We see them because of the tiny, staggered parallel lines required to reproduce an angled line using a grid pixel array. You can optimize to reduce this effect by using different anti-aliasing techniques that are specific to the type of display you're using. Apple tends to favor "soft" anti-aliasing techniques over accuracy. They also have the benefit of controlling the entire software stack, so their image processing can optimize for the IPS display used in iOS devices. Some people find the softness downright obnoxious, while others think it results in the most pleasing image. I prefer Apple's method. Color perception is a very broad topic. It's one of those "why is the sky blue" questions that we've all pondered (and frequently arrive at a false conclusion). I'm not nearly knowledgable enough to talk about it with authority, but you should definitely do some Googling, and be sure to keep an open mind. The answer is probably not what you think.
SomeCallMeTim 15y ago
I am pretty certain you're correct. Those Moire patterns were the ones you see when a fast image-scaling algorithm is taking every nth pixel, not what you should see with different underlying pixel layouts. In fact, looking at the iPhone zoom, it's blurred in the middle -- it's absolutely had a filter applied to it. You still see some Moire in the center of the image, but that could be a filter artifact. You'd need the chart rendered at exactly screen resolution, with proper filtering for that resolution (so that it doesn't end up with Moire patterns inherent in the image), and then to display it using code that's guaranteed not to apply a filter. He seems to be using the built-in slideshow apps in all cases (or maybe a browser?), and it looks like the iOS one has a better scaling filter. Which is interesting, but not what he's trying to prove. On top of all of that -- his photographic resolution can also be coming into play, adding more Moire patterns than you'd see in person. The lower resolution iPhone screen may not have had as much interference with his camera resolution. So I'd say it's an article fail.
smackfu 15y ago
Aren't these charts intended for analog cameras anyways?
- bradleyland 15y ago
  They're intended to be printed at high-resolution, then captured (photographed) at a distance where the fine lines at the center of the image greatly exceed the resolution capabilities of the capture device (usually a camera). With a film camera, you could test the limits of the lens and film. With a digital camera, you test the limits of the lens and sensor. The regularity of a pixel array in a camera sensor or display results in the Moire pattern you see in these test images.
bajsejohannes 15y ago
I thought the same, but it seems they display the image 1:1, so no scaling is applied. If that's the case, then I doubt any of the phones will process the image. (Though there might be reasons to do that, which I'm not aware of)
- bradleyland 15y ago
  The fact that the author references 600 dpi, then displays the image at 1:1 reveals that he really has no idea what he's dealing with. To further complicate matters, the author's source image is an already anti-aliased bitmap. Re-sizing this image in order to test appropriately on varying screen resolutions will result in anti-aliasing errors. The test images should be rendered at target resolutions from a vector based format like SVG. This is what the test image looks at zoomed to 400% with no anti-aliasing and no additional processing. This is a pixel-by-pixel resolution of the mess at the center of that image: http://i.imgur.com/WeoI4.png http://i.imgur.com/WeoI4.png When displayed 1:1 on a device, that mess will show up. DPI is dots per inch. This is an indication of resolution density. When you scale the display of an image on a fixed size screen, you also scale the DPI. That is to say, you could use a 1 million DPI image, and still, when displayed at 1:1 on a smartphone, the DPI of the resulting image is the same (the extra pixels are clipped outside the boundaries of the screen). Keep in mind that this test pattern should appear like a fractal. No matter how far you zoom in, the pattern will look the same. The points at the center should scale down to near infinite resolution. The resolution limitation is only that of the display device. A more suitable test that gets closest to the hardware would work something like this: * Start with an SVG test pattern * Record the native resolution of each device in pixels * Render the SVG test pattern with no anti-aliasing (or at least consistent anti-aliasing) at each target resolution * Display the image on the device using a custom developed application that disables or avoids any internal scaling or image processing capabilities of the software * Observe the results If you wanted to test the rendering capabilities of the entire software stack, you should still start with an SVG and use whatever built-in components will render SVG. Starting from a large bitmap tells us how well the bitmap scaling software works, but you shouldn't zoom to display at 1:1. You should allow the software to resize and see how well it scales. Each of these is another step removed from the core question asked in the article: Does the pentile screen impact image quality? To answer this question, we must rule out other factors, such as software rendering capabilities. The final error made is that of attempting to capture the results with a finite resolution capture device (a camera). At a pixel level, your capture would need to greatly exceed the resolution of the target display in order to examine the results at a pixel level or to reproduce the image in a way that objectively represents the perceived result to an audience not present at the time of testing.

sebbi 15y ago

Also the images don't appear to be fullscreen on any of the phones. If some highres image is being scaled and filtered to fit on the screen, than this "test" is definetly not an indicator of screen quality.

theatrus2 15y ago

The color test was also bunk - no deltaE tracking, no gamut comparisons, no brightness level comparisons.

yardie 15y ago

what we'd expect to see, at the price of accurate reproduction Can you explain what this means. Since our eyes are the only reference we have for interpreting colors (the other being wavelength, and that's just a number) I'm curious what an accurate color really means. I know that we can see billions of colors but only a thousands at the same time.

bradleyland 15y ago
I'm referring to the star chart, not the colors. You'll notice the curvy artifacts in the chart that looks like the ones found at this link: http://www.google.com/search?q=siemens+star+chart http://www.google.com/search?q=siemens+star+chart Those curves are an illusion. They don't actually appear in the image. We see them because of the tiny, staggered parallel lines required to reproduce an angled line using a grid pixel array. You can optimize to reduce this effect by using different anti-aliasing techniques that are specific to the type of display you're using. Apple tends to favor "soft" anti-aliasing techniques over accuracy. They also have the benefit of controlling the entire software stack, so their image processing can optimize for the IPS display used in iOS devices. Some people find the softness downright obnoxious, while others think it results in the most pleasing image. I prefer Apple's method. Color perception is a very broad topic. It's one of those "why is the sky blue" questions that we've all pondered (and frequently arrive at a false conclusion). I'm not nearly knowledgable enough to talk about it with authority, but you should definitely do some Googling, and be sure to keep an open mind. The answer is probably not what you think.

SomeCallMeTim 15y ago

I am pretty certain you're correct. Those Moire patterns were the ones you see when a fast image-scaling algorithm is taking every nth pixel, not what you should see with different underlying pixel layouts. In fact, looking at the iPhone zoom, it's blurred in the middle -- it's absolutely had a filter applied to it. You still see some Moire in the center of the image, but that could be a filter artifact. You'd need the chart rendered at exactly screen resolution, with proper filtering for that resolution (so that it doesn't end up with Moire patterns inherent in the image), and then to display it using code that's guaranteed not to apply a filter. He seems to be using the built-in slideshow apps in all cases (or maybe a browser?), and it looks like the iOS one has a better scaling filter. Which is interesting, but not what he's trying to prove. On top of all of that -- his photographic resolution can also be coming into play, adding more Moire patterns than you'd see in person. The lower resolution iPhone screen may not have had as much interference with his camera resolution. So I'd say it's an article fail.

smackfu 15y ago

Aren't these charts intended for analog cameras anyways?

bradleyland 15y ago
They're intended to be printed at high-resolution, then captured (photographed) at a distance where the fine lines at the center of the image greatly exceed the resolution capabilities of the capture device (usually a camera). With a film camera, you could test the limits of the lens and film. With a digital camera, you test the limits of the lens and sensor. The regularity of a pixel array in a camera sensor or display results in the Moire pattern you see in these test images.

bajsejohannes 15y ago

I thought the same, but it seems they display the image 1:1, so no scaling is applied. If that's the case, then I doubt any of the phones will process the image. (Though there might be reasons to do that, which I'm not aware of)

bradleyland 15y ago
The fact that the author references 600 dpi, then displays the image at 1:1 reveals that he really has no idea what he's dealing with. To further complicate matters, the author's source image is an already anti-aliased bitmap. Re-sizing this image in order to test appropriately on varying screen resolutions will result in anti-aliasing errors. The test images should be rendered at target resolutions from a vector based format like SVG. This is what the test image looks at zoomed to 400% with no anti-aliasing and no additional processing. This is a pixel-by-pixel resolution of the mess at the center of that image: http://i.imgur.com/WeoI4.png http://i.imgur.com/WeoI4.png When displayed 1:1 on a device, that mess will show up. DPI is dots per inch. This is an indication of resolution density. When you scale the display of an image on a fixed size screen, you also scale the DPI. That is to say, you could use a 1 million DPI image, and still, when displayed at 1:1 on a smartphone, the DPI of the resulting image is the same (the extra pixels are clipped outside the boundaries of the screen). Keep in mind that this test pattern should appear like a fractal. No matter how far you zoom in, the pattern will look the same. The points at the center should scale down to near infinite resolution. The resolution limitation is only that of the display device. A more suitable test that gets closest to the hardware would work something like this: * Start with an SVG test pattern * Record the native resolution of each device in pixels * Render the SVG test pattern with no anti-aliasing (or at least consistent anti-aliasing) at each target resolution * Display the image on the device using a custom developed application that disables or avoids any internal scaling or image processing capabilities of the software * Observe the results If you wanted to test the rendering capabilities of the entire software stack, you should still start with an SVG and use whatever built-in components will render SVG. Starting from a large bitmap tells us how well the bitmap scaling software works, but you shouldn't zoom to display at 1:1. You should allow the software to resize and see how well it scales. Each of these is another step removed from the core question asked in the article: Does the pentile screen impact image quality? To answer this question, we must rule out other factors, such as software rendering capabilities. The final error made is that of attempting to capture the results with a finite resolution capture device (a camera). At a pixel level, your capture would need to greatly exceed the resolution of the target display in order to examine the results at a pixel level or to reproduce the image in a way that objectively represents the perceived result to an audience not present at the time of testing.