Distortions

This page details what means were used to distort the included scenes. In all cases the dist_par values in the objective scores equal to the values needed to achieve the desired distortions (given the use of the same tools). The bits per pixel value for compression-based distortions can also be found in the objective scores, as bpp_actual.

Compression-based distortions

Image encoders

  • JPEG/JPEG 2000 - Compressed the individual perspectives with MATLABs imwrite. In all cases the dist_par value was set as the 'Quality' parameter.

  • BPG - Compressed the individual perspectives with the BPG executable. Parameter -m was set to 1. The dist_par value equals to a scaled -q parameter. The scaling equation is q=50-(51/200)*dist_par.

Video encoders

FFMPEG was used to compress all the perspectives in a serpentine scan order. In all cases the -crf parameter was used to determine the output quality. The dist_par was scaled:

  • VP9 and AV1 --> crf=(63*(100-dist_par))/100

  • AVC and HEVC --> crf=(51*(100-dist_par))/100

Non-compression-based distortions

Gaussian and impulse noise

In both cases MATLABs imnoise function was used. All noise was applied uniformly across the perspectives.

  • Gaussian noise - The dist_par values were used to set the mean (m) and variance (var_gauss) of the Gaussian noise. The dist_par value includes both the m and var_gauss values, separated by an underscore (m_var_gauss)

  • Impulse noise - The dist_par values were used to set noise density. As such, dist_par was directly equal to the d parameter in imnoise.

Unsharp mask

MATLABs imsharpen function was used for the application of unsharp mask. Each perspective was sharpened with the same parameters.

  • The dist_par values set the Radius (the standard deviation of the Gaussian lowpass filter) and the Amount (the strength of the sharpening effect) parameters of the imsharpen function. Similarly to Gaussian noise, the dist_par includes both values separated by an underscore (Radius_Amount).

Geometric distortion

The lensdistort function was used to geometrically distort the individual perspectives.

  • Barrel distortion - The dist_par sets the k parameter in the lensdistort function.

  • Pincushion distortion - The dist_par is equal to the negative value of the k parameter in the lensdistort function.