Investigating symmetric and asymmetric stereoscopic compression using the PSNR image quality metric

Context: In this paper we will assess the two different stereoscopic image compression methods, symmetric and asymmetric encoding. As stereoscopic images require an increase in storage space compared to monoscopic images, the compression of stereo images is one of the most important factors to enable the extensive use of three-dimensional systems. Substantial research effort has been focused on digital image compression using JPEG, however, the question of how much compression to apply and the most appropriate compression method to use still remains unresolved. Based on theories of binocular suppression, it is often contended that the binocular percept of a stereo image pair is dominated by the high quality component. Thus, when one image of a stereo pair is compressed, such that a high quality is maintained, the other view may be coded more heavily without introducing visible artefacts into the binocular percept. Method: Symmetric encoding of the image pair is when the left and right images are compressed by equal amounts resulting in equal degradation. Asymmetric encoding is when the compression and therefore degradation of the left and right images is unequal. The question of whether to compress one eye´s view heavily (asymmetric) or both with equal amounts (symmetric) still remains unanswered. We investigate symmetric and asymmetric encoding of compressed images using Peak Signal to Noise Ratio (PSNR). Photographic, computer generated photo realistic and computer generated non-photo realistic test images are compressed, symmetrically and asymmetrically, using JPEG and, in each case a constant file size is maintained between the pairs. Results: The PSNR results from this investigation showed that in all cases, symmetric encoding, as opposed to asymmetric encoding, produced significantly better results. Conclusion: We conclude that in general for stereoscopic image compression, using JPEG, a symmetric as opposed to asymmetric compression approach across the left - - and right images should be used.