Histogram (hIST)

• Status:   PNG Specification
• Location:   after PLTE, before first IDAT
• Multiple:   no

A histogram is nothing more than a frequency-of-occurrence table, and the PNG hIST chunk gives the approximate frequencies of occurrence for pixels of various colors. This information is typically used to decide which colors are the most important if the system is not capable of displaying all of them. Rather than force the decoder to count pixels every time the image is displayed, the histogram places the burden on the encoder, which performs the task only once.

PNG's hIST implementation is tied to the PLTE chunk; if there is no palette, hIST is not allowed. This and one or two other limitations were later recognized and addressed by the sPLT chunk, which we'll discuss next; it is generally favored over hIST, but the latter is smaller, and either may be used. The histogram must contain exactly as many entries as PLTE contains, and each entry is a 16-bit unsigned integer. Since such integers can only represent numbers in the range 0-65,535 and there may be millions of pixels of a given color, the histogram entries often must be scaled and are therefore inexact. The sole exception is the value zero; it is guaranteed to mean that there are no pixels of the corresponding color. A nonzero count that would otherwise be scaled and rounded to zero must instead be rounded up to one.

Truecolor images that include a PLTE chunk as a suggested quantization are a special case. The histogram counts are dependent on the algorithm used by the encoder for quantizing the pixels; if the decoder happens to use a different algorithm, its counts would be different, too. The upshot is that the histogram is particularly approximate in this case. Because truecolor images typically have far more colors than palette entries, the palette entries that do appear should always represent at least one pixel; thus there should be no zero counts in the histogram.