SIAT VIDEO TEAM

SIAT Picture-level Just Noticeable Difference-based Stereoscopic Image Database

Introduction

We study the Picture-level Just Noticeable Difference (PJND) of symmetrically and asymmetrically compressed stereoscopic images for JPEG2000 and H.265 intra coding. We conduct interactive subjective quality assessment tests to determine the PJND point using both a pristine image and a distorted image as a reference. We generate two PJND-based stereo image datasets, including Shenzhen Institutes of Advanced Technology-picture-level Just noticeable difference-based Symmetric Stereo Image dataset (SIAT-JSSI) and Shenzhen Institutes of Advanced Technology-picture-level Just noticeable difference-based Asymmetric Stereo Image dataset (SIAT-JASI). Each dataset includes ten source images, respectively. The PJNDPRI and PJNDDRI are provided. PJNDPRI reveals the minimum distortion against a pristine image. PJNDDRI reveals the minimum distortion against a distorted image.

Keywords: Just Noticeable Difference (JND), Stereoscopic image, Image quality assessment, Subjective quality assessment.

Publications

[1] Chunling Fan, Yun Zhang, Raouf Hamzaoui, and Qingshan Jiang, Interactive Subjective Study on Picture-level Just Noticeable Difference of Compressed Stereoscopic Images, in: IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), 2019. (pdf , poster)

[2] Chunling Fan, Yun Zhang, Huan Zhang, Raouf Hamzaoui, and Qingshan Jiang, Picture-level Just Noticeable Difference for Symmetrically and Asymmetrically Compressed Stereoscopic Images: Subjective Quality Assessment Study and Datasets. Vol. 62, pp.140-151, 2019. (pdf , Database@ IEEE Dataport)

Experimental settings

For symmetric compression, the same compression parameter was used in the left view and right view, respectively. For asymmetric compression, different compression parameters were used in the left view and right view, respectively. HM-16.7-00 and FFMpeg-n3.2.8 for standard H.265 were adopted to encode each view of the stereoscopic images in intra-frame mode separately. FFMpeg-n3.2.8 was used to encode/decode between BMP and YUV. HM-16.7-00 was used to encode YUV into bit stream. The QPs were varied from 1 to 51 for both the left and right views. For JPEG2000 compression, we used MATLAB code “imwrite (image, imageName, ‘CompressionRatio’, CR)”, and CR ranges from 1 to 300.

The experiment is conducted in two parts. Experimental settings are shown in Table 1. There were a total of 50 non-expert subjects participated in the experiment and each test stereoscopic image was evaluated by 35 subjects. The quality scores assigned by each subject should be normalized per person before computing PJND. All subjects taking part in the experiment are graduate students. Every subject had been told the procedure of the test and had watched a training sequence before starting the experiment.

Fig.1. Ten stereoscopic sequences from the NAMA3DS1 database which are used in the subjective experiment.