新的全参考音视频同步感知质量评价模型

提出一种利用协惯量分析构建的全参考音视频同步感知质量模型。通过对齐得到待测音频与视频的同步误差。将音视频内容分为纯净语音、无语音和有背景语音3类。将纯净语音类分为视频中有说话人和无说话人2个子类。分别对各类选取多维特征,利用协惯量分析从特征中获得音视频最相关的特征映射和相关程度。通过参考音视频得到相关程度曲线并得到同步误差到感知质量的映射关系。结果表明该模型评测结果与主观实验结果有较好相关性。

Novel full reference perceptual quality metric for audio-visual asynchrony

A full reference model was proposed to evaluate the perceptual quality of audiovisual asynchrony.A standard synchronization process was used to determine the time difference between audio and video.The mapping between the time difference and the perceptual quality was derived by co-inertia analysis.The co-inertia analysis extracted the most related component from audio and video features,and then formed a mapping for each audiovisual sequence.Audiovisual contents were divided into three categories: clean speech,non speech and mixed speech.The clean speech category was further split into two subcategories.Audio and video features were chosen separately for each category.Subjective test results showed that the proposed model conforms well with subjective results.