Influence of the source content and encoding configuration on the perceived quality for scalable video coding

Abstract

In video coding, it is commonly accepted that the encoding parameters such as the quantization step-size have an influence on the perceived quality. It is also sometimes accepted that using given encoding parameters, the perceived quality does not change significantly according to the encoded source content. In this paper, we present the outcomes of two video subjective quality assessment experiments in the context of Scalable Video Coding. We encoded a large set of video sequences under a group of constant quality scenarios based on two spatially scalable layers. One first experiment explores of the relation between a wide range of quantization parameters for each layer and the perceived quality, while the second experiment uses a subset of the encoding scenarios on a large number of video sequences. The two experiments are aligned on a common scale using a set of shared processed video sequences, resulting in a database containing the subjective scores for 60 different sources combined with 20 SVC scenarios. We propose a detailed analysis of the experimental results of the two experiments, bringing a clear insight of the relation between the encoding parameters combination of the scalable layers and the perceived quality, as well as spreading light on the differences in terms of quality depending on the encoded source content. As an endeavour to analyse these differences, we propose a classification of the sources with regards to their relative behaviour when compared to the average of other source contents. We use this classification to identify potential factors to explain the differences between source contents.

Content-aware objective video quality assessment

“…36 Therefore, it would be more desirable as an S-shape function such as (e) logistic, (j) complementary error, or (k) complementary cumulative raised cosine to take into account the saturation effect of human vision. Note that, some public video quality databases includes only four distortion levels per scene (e.g., IRCCyN IVC Influence Content 18 ), which greatly limits the number of data points available in the training process. That is, due to limitations in the current available data, it is inconvenient to model the relationship VQM → DMOS with more than two parameters.…”

“…That is, we compare between the following methods PSNR, SSIM, SOVQM, VQAD, and CPSNR on the IRCCyN IVC Influence Content database 18 as well as on the CIF and 4CIF EPFL-PoliMI Video Quality Assessment Databases. 28 The performance appraised on the IRCCyN IVC Influence Content database is high for most of the tested VQMs except the SSIM (cf.…”

“…In these databases, each source sequence was compressed to generate a set of processed sequences using H.264 compression standard 18,23,29 Note that the CIF and the 4CIF EPFL-PoliMI Video Quality Assessment Databases 28 have videos compressed with H.264 followed by packet loss simulation, i.e., this database allows us to explore the behavior of the proposed methodology under different distortion types. Each processed sequence has its DMOS (we use DMOS because it measures the change in quality between two versions of the same stimulus with minimal impact on estimating model performance 43 ) value obtained through subjective experiments.…”

“…Since the IRCCyN and IVP databases were used during the content related indexes selection, we use another three sets of data to validate our methodology with the purpose of avoiding cross-validation errors. In particular, we use the IRCCyN IVC Influence Content database 18 as well as the CIF and 4CIF EPFL-PoliMI Video Quality Assessment Databases. 28 Note that these databases were not used in the entire process of model training, content related indexes and/or mapping functions selection, i.e., we use in this work a training, validation, and test sets to measure the performance of the proposed methodology.…”

“…The latter problem is mainly due to the strong dependency of VQMs on the video content. [12][13][14][15][16][17][18][19] Despite this dependency being well known, only a few existing quality measures directly account for the effects of content. For instance, Feghali et al, 12 Garcia et al, 17 Korhonen and You, 16 and Ou et al 19 proposed models that combine content related indexes, peak signal to noise ratio (PSNR), bit rate, spatial and temporal resolution for estimating the quality of compressed video sequences.…”