Is this kind of strategy used in many real-world image compression algorithms?
billiam
This doesn't seem like a very accurate compression. There would need to be some other kind of information that goes along with Y'CrCb to get a truly accurate representation of the original image right? Sure the reconstructed result "looks" pretty good, but if I was playing a video game and every texture looked approximated like the one above I wouldn't be too happy. Is there some other data that is used along with Y'CrCb for the compression? Perhaps the use isn't for video games?
smohidee
This definitely isn't a lossless compression algorithm due to the downsampling, but perhaps you could tweak the ratio to get a closer approximation to the original. According to Wikipedia there is some perceptual basis for emphasizing lightness data over color data (likely due to the different numbers of rods and cones in the eyes). This specific technique is called chroma subsampling and is apparently used in real encoding schemes like JPEG: https://en.wikipedia.org/wiki/Chroma_subsampling
motoole2
There are definitely ways to improve the compression algorithm shown here. Generally, downsampling (and upsampling) an image is a really bad way to compress visual data because high-frequency information is completely lost. But this still works reasonably well when done to only the chroma channel, since the human visual system is less sensitive to changes in chromatic information when compared to luminance. As mentioned by @smohidee , real compression techniques can exploit this to get better image compression performance.
Is this kind of strategy used in many real-world image compression algorithms?
This doesn't seem like a very accurate compression. There would need to be some other kind of information that goes along with Y'CrCb to get a truly accurate representation of the original image right? Sure the reconstructed result "looks" pretty good, but if I was playing a video game and every texture looked approximated like the one above I wouldn't be too happy. Is there some other data that is used along with Y'CrCb for the compression? Perhaps the use isn't for video games?
This definitely isn't a lossless compression algorithm due to the downsampling, but perhaps you could tweak the ratio to get a closer approximation to the original. According to Wikipedia there is some perceptual basis for emphasizing lightness data over color data (likely due to the different numbers of rods and cones in the eyes). This specific technique is called chroma subsampling and is apparently used in real encoding schemes like JPEG: https://en.wikipedia.org/wiki/Chroma_subsampling
There are definitely ways to improve the compression algorithm shown here. Generally, downsampling (and upsampling) an image is a really bad way to compress visual data because high-frequency information is completely lost. But this still works reasonably well when done to only the chroma channel, since the human visual system is less sensitive to changes in chromatic information when compared to luminance. As mentioned by @smohidee , real compression techniques can exploit this to get better image compression performance.