Content-based bit stream layer video quality assessment model

Abstract

The invention discloses a content-based bit stream layer video quality assessment model, and belongs to the technical field of multi-media. A compressed domain model can directly extract a video parameter from a code stream, for performing non-invasive quality monitoring in real time. The model mainly aims at an H.264 video coding standard, for assessing the quality of coding distortion. A basic relation is built between sensed quality and a quantization parameter (QP). In view of significant dependence of the sensed quality on video content, the video content is defined to be combinations with different time complexity and spatial complexity. According to the model provided by the invention, the occupied ratio (SPM_Ratio) of a small prediction bock and a proportion (Ave_Coef) of an average DCT nonzero number in each 4*4 block in an I frame are used for synchronously describing the spatial complexity of the video. An average normalization parameter (Ave_Mvlen) of a suffix length of motion vector information in the code stream is used for describing the spatial complexity of the video. All information can be directly extracted from the code stream, with no need to perform decoding operation.

Description

technical field [0001] The invention proposes a video quality evaluation model based on a compressed domain, which belongs to the technical field of multimedia. Background technique [0002] In recent years, with the vigorous development of multimedia technology, network video services, videophone, IPTV, etc. have gained obvious popularity in our daily life. However, the quality of these applications often cannot be monitored and guaranteed in real time. Therefore, it is very necessary to establish an objective model for network video quality assessment. [0003] From the perspective of whether original video information is needed, video quality evaluation can be divided into subjective evaluation and objective evaluation. Subjective evaluation requires the subjects to observe a series of tested videos in a specific environment and score according to the pre-specified scoring standards. The main methods are: DSIS (Double Stimulus Impairment Scale) method, DSCQS (Double St...

AI Technical Summary

Problems solved by technology

However, deep learning requires a large amount of video data, and commonly used databases cannot reach this order of magnitude

In addition, there is still no complete scientific theory to support the feasibility of deep learning, so deep learning has not gained more recognition and achievements in the direction of video quality evaluation

Images