OpenCV data types

Question

depth Pixel depth in bits. The supported depths are:

IPL_DEPTH_8U Unsigned 8-bit integer
IPL_DEPTH_8S Signed 8-bit integer
IPL_DEPTH_16U Unsigned 16-bit integer
IPL_DEPTH_16S Signed 16-bit integer
IPL_DEPTH_32S Signed 32-bit integer
IPL_DEPTH_32F Single-precision floating point
IPL_DEPTH_64F Double-precision floating point

1. What these value actually stands for?

2. How much bits presents each one?

3. What is the difference between:

   Unsigned 8-bit integer and Signed 8-bit integer ?

   Unsigned 16-bit integer and Signed 16-bit integer ?

   If they demands 8 and 16 bit respectivly?

4. What's the sense to use data types with floating point?

Answer 1

An unsigned 8 bit has values from 0 to 255, while a signed 8 bit has values from -127 to 127. Most digital cameras use unsigned data. Signed data is mainly the result of an operation on an image, such as a Canny edge detection.

The reason for higher bit depth images, such as 16 bit, is more detail in the image. This allows more operations, such as white balancing or brightening the image, without creating artifacts in the image. For example, a dark image that has been brightened to much has distinct banding in the image. A 16 bit image will allow the image to be brightened more than an 8 bit image, because there is more information to start with.

Some operations work better with floating point data. For example, a FFT(Fast Fourier Transform). If too many operations are done on an image, then the error from rounding the pixel values to an integer every time, start to accumulate. Using a floating point number mitigates this, but doesn't eliminate this.