Automatic video comparison of the output of a video decoder
The automatic video comparison system for measuring the quality of decoded data described herein provides a method for measuring the quality of decoded data at the level of sub-units of a unit of data, for instance at the level of sub-blocks of a video frame. The system can therefore locate defects that may not otherwise be detected by an automated system that measures quality at the level of the entire frame. Processing encoded media is computationally intensive, thus the automatic video comparison system uses a distributed computing system in order to distribute the computations across many compute resources that are capable of operating in parallel.
1. A method for automatic detection of the quality of a decoded video stream, the method comprising:
receiving an encoded video stream;
decoding the encoded video stream into a decoded video stream, the decoded video stream comprising at least one decoded video frame;
producing a reference video data stream from the encoded video stream, the reference video data stream comprising at least one window of reference video data, the at least one window of reference video data comprising corresponding blocks of reference video data frames;
comparing the at least one decoded video frame with the at least one window of reference video data by comparing blocks of each of the at least one decoded video frame with the corresponding blocks of reference video data frames from the at least one window of reference video data, with the at least one decoded video frame and the blocks of reference video data frames having misaligned synchronization;
producing a quality measurement for each block of the at least one decoded video frame using a quality metric to indicate the amount of misaligned synchronization; and
determining for each block of the at least one decoded video frame the quality metric by indicating which of the at least one window of reference video data in the frame has a comparatively best quality measurement,
wherein the best quality measurement comprises a measured vector where a block of the frame that has the largest Peak Signal-To-Noise Ratio (PSNR) is the smallest among the reference video frames.
2. The method of claim 1 , wherein determining, for each decoded video frame, which reference video data frame from the at least one window of reference video data has comparatively the best quality measurements in accordance with the quality metric M is mathematically expressed as:
M
=
min
W
w
=
1
max
K
k
=
1
(
PSNR
k
)
for
K
components
and
a
comparison
window
of
W
.
3. The method of claim 1 , comprising determining the quality metric for each block of the at least one decoded video frame is performed without use of metadata from the decoded video frame.
4. The method of claim 1 , comprising producing the reference video data stream by decoding the encoded video stream.
5. The method of claim 1 , comprising producing the reference video data stream by extracting prediction information from the encoded data stream.
6. The method of claim 1 , wherein the step of comparing is used to determine the amount of the synchronization misalignment of the at least one decoded video frame and the blocks of reference video data frames.
7. The method of claim 1 , wherein the comparison is made from the decoded video frame data without use of extracted information about an image pixel sequence of the decoded video frame data.