Sensors produce values that vary randomly, causing unnatural colour reproduction. Reducing this variability increases the fidelity of the image, and improves other characteristics, such as compressibility.
Structure from motion
By tracking feature points in a sequence of images, the motion of the camera and the shape of objects in a scene may be computed.
Finding the correspondence of image patches in consecutive images allows for processing of pixels along the time dimension.
A modified value for each pixel may be computed by considering its value relative to it's neighbours.
Filters may be applied in the spatial and/or temporal domains to reduce noise and improve image appearance. Frequency domain design may be used to optimize filter performance.
Colour space transforms
Computations on pixels are best done in a colour space that meets a specific need, such as perceptual or statistical.