1. The principle of compression of the JPEG is based on splitting the image into squares of 8×8 pixels and fill those squares gradients. Suitable for photographs (where, by definition, a lot of gradients and little clear objects) and is not suitable for line art images of type schemes (for clarity, you should try to imagine getting, say, a single-pixel line when trying to represent it as a gradient of 8×8; it is also useful to consider this JPEG image at high magnification).
2. The principle of compression of GIF and PNG-8 (not to be confused with PNG-24) is based on narrowing the color palette to a minimum, sufficient for more or less correct display of the image. By narrowing the palette, each color can be encoded by fewer bytes, this (in addition to algorithms archiving) and achieved compression. The maximum number of shades is 256. Accordingly, GIF and PNG-8 are well suited for subjectively lossless (encoded each pixel) compression of line images with solid areas and a small number of colors (logos, graphs, charts, etc.). and is NOT suitable for the vast majority of photographic images (photographs of a number of shades are usually much larger than 256, if not more, the size of the file is unnecessarily large, since the GIF/PNG-8 honestly encode each pixel of the image).
3. PNG-24 (not to be confused with PNG-8) compression is generally lossless. In fact, specialized archiver for images. Accordingly, the volume of the output file most and usually much larger than GIF/PNG-8, and JPEG. Usually applied or, if necessary, lossless compression, or if you want to use transparency, either when you save the long, monotonous areas and/or unidirectional gradients (such gradients compress more efficiently than GIF/PNG-8).