Member Rara Avis
Don't be too quick to blame the software, guys. File compression, overly simplified, is based on finding binary similarities within the file. Got a big, 4K blue sky in your image? Instead of saving a byte for every sky pixel, I'll just say that "blue sky" is now called xyz and save the file with a 4096xyz tag - saving almost 4K. When the file is loaded by a program that understands the compression, the 4096xyz tag will be converted back to a nice pretty blue sky before being displayed on screen.
Some images, however, just don't have a whole lot of binary similarities and can be a nightmare to compress below a certain point. The algorithm is looking for groups it can replace with smaller definitions; in our blue sky example, we have to have a definition (saying xyz=0000ff, for example) and a declaration (our 4096xyz). Altogether, we might say it takes 16 bytes to define and declare our 4K blue sky. A big savings! But what if the largest group of similarities in your images is a string of 20 red pixels? Well, we only saved 4 bytes. Ouch. So a "busy" image, with a lot of switches from one color to another, just isn't going to compress very well. And, ironically, that is more likely to happen with a smaller photo or a low resolution. Sometimes, the only solution is to find another, more amenable image.
And, yes, Stephanos, I "could" raise the upload limit above 5K. But if your 16K photo is one of those that simply cannot be further compressed, that 16K is going to eventually be multiplied by every originating post you make, and then again by every person who views it. Opening a thread and waiting for a 5K photo (and most are actually 2-3K) takes a little more time than one with no picture, but only a little. And it's worth it to be able to "see" the person face-to-face. But larger images costs our Members more time, and Ceres more workload, and become less and less worth it. The 5K limit was a compromise that seems to work "most" of the time.
Now, would anyone like to know why jpg (unlike gif and zip) is considered a "lossy" compression scheme?