No reference metrics

Our two no reference metrics are dedicated to specific codecs: the first one metric is designed for MPEG-2 whereas the second one is designed for H.264.

For each codec, the corresponding metric measures the distortions generated by the encoding:

• No Reference metric for MPEG-2 video: measurement of blockiness visibility and blur perception.
• No Reference metric for H.264 (MPEG-4/AVC) video: measurement of objects sharpness and blur perception.

This is essential and makes a great difference with some other existing video quality measurement solutions. Indeed, some of these solutions are designed for a given encoding format. For example, many solutions are designed for MPEG-2 because they measure the visibility of MPEG-2 artefacts like blockiness and blur (please note that measuring blockiness on MPEG-2 is quite easy since all the blocks have a known location and size). And yet, these solutions are also proposed to measure the quality of videos that have been encoded using H.264 (MPEG-4/AVC). However :

• using H.264 encoding, the blocks have varied sizes and don't have previously defined locations (so, what is the point in measuring blockiness at incorrect locations?)
• the H.264 decoder contains a deblocking filter (so, what is the point of measuring blockiness on a video that has been processed by a deblocking filter?)
• the main artefact on H.264 encoded videos is not blockiness but is more detail loss (so, what is the point of considering blockiness as the major artefact to determine video quality).

Therefore, one has to be conscient that a video quality metric designed for MPEG-2 can't measure the quality of videos encoded with H.264.