What is the subpixel?
Surface array camera imaging for the smallest unit in pixels.
A CMOS camera chip, for example, the pixel spacing is 5.
Camera, will be continuous in the physical world image processing.
Each pixel to the imaging plane. Only on behalf of its nearby color.
As for the 'near' to what degree?
Is very difficult to explain.
There are 5 between two pixels.
2 micron distance, on the macro can be thought of as together.
But in the micro level, there are infinite more small things exist between them.
The more small things we call it 'pixel'.
'Pixel' should be actually exist, only hardware not a tiny sensors detect it.
So put it on the software approximately calculated.
Subpixel accuracy of subpixel accuracy refers to the segment between two adjacent pixels.
The input values are usually half, a third or a quarter.
This means that each pixel will be divided into smaller units to implement interpolation algorithm for these smaller units.
For example, if you choose a quarter, is equivalent to each pixel on the horizontal and vertical are treated as four pixels to calculate.
Therefore, if a 5 x5 pixel image pixel accuracy, and chose a quarter is to create a 16 x16 discrete lattice, and then the interpolation of the lattice.
Please refer to the below, the red dot represents the original pixels, pixel point and black dot represents the new produce.
The pixels in the application of machine vision in machine vision, subpixel is a common concept, in a lot of function, can choose whether to use the pixels, in the measurement, such as position, straight line, round, will appear the pixels.
Such as measuring a circle diameter of 100.
The behind of 0.
12 are the pixels.
Because from the pixel can understand that the smallest physical unit is pixel industrial camera, but we will still get the decimal point in the machine vision measurement values, this is calculated by the software, actually in the real cases, may not be very accurate.
This value is usually easier to reaction in grayscale, in binary image, because the value is 0, 1.
So many of the functions will not necessarily subpixel calculation.