Image is a scalar field one or more channels. The channels are intensity fields that can be 8,16,32 or 64 bit. The 8 bit is an integer between 0 and 255, 16 is an integer between 0 and 2^16-1 = 65,532, and 32 and 64 are floating point representation with either single or double precision. The floating point representation allows positive and negative values and fractions. Single precision is 32 bits/4 bytes and gives around 8 digits of accuracy. Double precision is twice the size and gives you around 16 digits of accuracy. Floating point is typically only used when you perform calculations on images such as ratios or when you compute the difference between two images.
The Image variable can also be used outside of the typical use case for images and where you have a uniformly discretized field of numbers. For example if you are computing PDE solutions. In that case you typically stay in double precision.
The image is defined on a grid with the origin being in the lower left corner. This is different than is typical in image processing programs like ImageJ where the origin is in the upper left corner. This is done to make it easier to mix the image and the derived quantities like thresholds and polygons.