Design for Embedded Image Processing on FPGAs, 2nd Edition

12Image and Video Coding

The information content within images and video may be compressed to reduce the volume of data for storage, reduce the bandwidth required to transmit data from one point to another (for example over a network), or minimise the power required by the system (both for compression and data transmission). Compression is only possible because images contain significant redundant information. There are at least four types of redundancy within images that can be exploited for compression:

• Spatial redundancy results from the high correlation between adjacent pixels. Since adjacent pixels are likely to come from the same object, they are likely to have similar pixel value or colour.
• Temporal redundancy comes from the high correlation between the frames of a video sequence. Successive frames tend to be very similar, especially if there is limited movement within the sequence. If any movement can be estimated, then the correlation can be increased further by compensating for the motion.
• Spectral redundancy reflects the correlation between the colour components of a standard RGB image, especially when looking at natural scenes. Within hyperspectral images, there is a strong correlation between adjacent spectral bands.
• Psychovisual redundancy results from the fact that the human visual system has limited spatial, temporal, and intensity resolution. It is tolerant of errors or noise, especially within textured regions.

A codec (coder–decoder) can be categorised ...