Video quality assessment method and apparatus

Patent metadata

Field	Value
Publication number	US-11070794-B2
Application number	US-201916689742-A
Country	US
Kind code	B2
Filing date	Nov 20, 2019
Priority date	Nov 21, 2018
Publication date	Jul 20, 2021
Grant date	Jul 20, 2021

Abstract

In a video quality assessment method, an assessment model is first generated based on a subjective assessment result of a user on each sample in a sample set and based on a parameter set (a parameter type in the parameter set may include at least one of a packet loss rate, a delay, and a jitter) of each sample. Therefore, when video quality is being assessed, a parameter set of a to-be-assessed video is obtained first, where the parameter set of the to-be-assessed video has a same parameter type as the parameter set of each sample that is used to generate the assessment model; and then video quality of the to-be-assessed video is assessed based on the assessment model and the parameter set of the to-be-assessed video, to obtain an assessment result.

First claim

What is claimed is:

1. A method comprising:
obtaining a video parameter set of a video;
generating an assessment model, wherein the assessment model is based on a subjective assessment result of a user on each sample in a sample set and based on a sample parameter set of each sample, wherein the video parameter set and the sample parameter set have a same parameter type, and wherein the parameter type comprises at least one of a jitter, a delay, or a packet loss rate, wherein the assessment model is generated by:
classifying N samples into a test set and a plurality of training sets; wherein the N samples correspond to N test videos, N an integer greater than or equal to 2;
classifying samples in each of the plurality of training sets based on the different parameter types in the parameter set, to generate a plurality of initial decision tree models, wherein the plurality of initial decision tree models comprise an initial decision tree model corresponding to each of the plurality of training sets;
testing the plurality of initial decision tree models based on at least one sample in the test set to obtain a plurality of groups of test results, wherein the plurality of groups of test results have a one-to-one correspondence with the plurality of initial decision tree models and each group of test results comprise at least one test result that has a one-to-one correspondence with the at least one sample; and
determining a decision tree model from the plurality of initial decision tree models as the assessment model based on the plurality of groups of test results, wherein the decision tree model is an initial decision tree model corresponding to a first group of test results, and a test result corresponding to each sample in the first group of test results is the same as a subjective assessment result of the sample; and

assessing a video quality of the video using the assessment model and the video parameter set to obtain an assessment result.

2. The method of claim 1 , wherein before obtaining the assessment model, the method further comprises:
obtaining the N test videos; and
obtaining a sample corresponding to each of the N test videos to obtain the N samples, wherein the sample comprises the sample parameter set and the subjective assessment result.

3. The method of claim 2 , further comprising further obtaining the N test videos by adding an impairment corresponding to each parameter type to a source video, wherein the impairment is represented by a parameter value, and wherein parameter values corresponding to all the parameter types form the sample parameter set.

4. The method of claim 3 , further comprising:
determining a delay value of a first test video of the N test videos based on a sending time and a receiving time of a data packet of the first test video;
determining a first quantity of data packets of the first test video that are sent in a unit time;
determining a second quantity of data packets of the first test video that are received in the unit time;
calculating a packet loss rate value of the first test video as a ratio of the second quantity to the first quantity;
determining a jitter value of the first test video based on a time interval between a sending time of each of the data packets in the unit time and a receiving time of each of the data packets in the unit time; and
establishing a mapping relationship between the subjective assessment result and at least one of the delay value, the packet loss rate value, or the jitter value to obtain a first sample corresponding to the first test video.

5. The method of claim 1 , further comprising:
calculating an information gain of each parameter type in M samples of the training set;
determining that a first parameter type corresponding to a maximum value of a plurality of calculated information gains is a root node, wherein the M samples make up all samples in the training set, wherein the information gain of each parameter type in the M samples is a difference between first information entropy and second information entropy, wherein the first information entropy is based on a value of a subjective assessment result of each sample in the M samples used as a whole, wherein the second information entropy is a sum of first sub information entropy and second sub information entropy, wherein the first sub information entropy is based on values of subjective assessment results of L samples, wherein the second sub information entropy is based on values of subjective assessment results of P samples, wherein the L samples and the P samples are based on dividing the M samples based on a division condition that a value of the parameter type is greater than or equal to a preset parameter value, wherein M=L+P, and wherein M, L, and P are positive integers;
dividing the M samples into two level-1 subsets based on a division condition that a value of the root node is greater than or equal to a value of the first parameter type, wherein the value of the first parameter type corresponds to a maximum value of information gains of the first parameter type in the M samples;
determining, based on information gains of each parameter type in the two level-1 subsets, that a second parameter type corresponding to a maximum value of a plurality of information gains corresponding to each level-1 subset is a subnode of the level-1 subset;
dividing at least two samples in the level-1 subset into two level-2 subsets based on a division condition that a value of the subnode is greater than or equal to a value of the second parameter type, wherein the value of the second parameter type corresponds to a maximum value of information gains of the second parameter type in the level-1 subset; and
obtaining the initial decision tree model when at least two subjective assessment results of at least two samples comprised in any one of the two level-2 subsets are the same or any one of the two level-2 subsets comprises only one sample.

6. The method of claim 1 , wherein a server implements the method, and wherein the method further comprises receiving the subjective assessment result from a terminal.

7. An apparatus comprising:
a memory configured to store instructions; and
a processor coupled to the memory and configured to execute the instructions to:
obtain a video parameter set of a video;
generate an assessment model, wherein the assessment model is based on a subjective assessment result of a user on each sample in a sample set and based on a sample parameter set of each sample, wherein the video parameter set has and the sample parameter set have a same parameter type, and wherein the parameter type comprises at least one of a jitter, a delay, or a packet loss rate, wherein the assessment model is generated by:
classifying N samples into a test set and a plurality of training sets; wherein the N samples correspond to N test videos, N is an integer greater than or equal to 2;
classifying samples in each of the plurality of training sets based on the different parameter types in the parameter set, to generate a plurality of initial decision tree models wherein the plurality of initial decision tree models comprise an initial decision tree model corresponding to each of the plurality of training sets;
testing the plurality of initial decision tree models based on at least one sample in the test set to obtain a plurality of groups of test results, wherein the plurality of groups of test results have a one-to-one correspondence with the plurality of initial decision tree models and each group of test results comprise at least one test result that has a one-to-one correspondence with the at least one sample; and

determining a decision tree model from the plurality of initial decision tree models as the assessment model base

Assignees

• Huawei Tech Co Ltd

Inventors

• Hong Hanshu
• Dong Feng
• Song Wei
• Yang Qingping

Classifications

• G06N5/01

  Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound · CPC title

• H04N19/154

  Measured or subjectively estimated visual quality after decoding, e.g. measurement of distortion (use of rate-distortion criteria H04N19/147) · CPC title

• H04N21/44008

  involving operations for analysing video streams, e.g. detecting features or characteristics in the video stream (arrangements characterised by components specially adapted for monitoring, identification or recognition of video in broadcast systems H04H60/59) · CPC title

• H04N21/44209

  Monitoring of downstream path of the transmission network originating from a server, e.g. bandwidth variations of a wireless network (arrangements for maintenance or administration in data switching networks involving bandwidth and capacity management H04L41/0896) · CPC title

• H04N17/00 Primary

  Diagnosis, testing or measuring for television systems or their details · CPC title