12704_SCTE_Broadband_Nov2016_COMPLETE_lowres

technical

dimension. Each physical dimension corresponds to a specific spatial frequency. The plurality of receptive fields of a certain dimension results in higher contrast sensitivity to the related spatial frequency. There are many existing models that explain the response of the HVS to the light spatial frequencies. Figure 4 below shows the Barten model.

temporal Contrast Sensitivity Function (CSF), we can more accurately present the sensitivity to each pixel distortion by considering the amount of their changes over time. By weighing the importance of these sensitivities against less relevant elements in a video frame, it’s possible to measure which distortion is visible to the visual system and which is not. Correct measurement of visible distortion allows the encoder to determine the highest level of compression of each pixel, image and video scene to achieve the desired video quality. This balance between video quality and bandwidth is beyond the capabilities of competing encoding techniques, such as CBR or CVBR encoding. Harmonic has spent extensive R&D resources on this area and reached the conclusion that taking into account the HVS is not enough to achieve true constant quality. While science has made great progress in understanding the HVS, there are still some pieces that are not known. The challenge is building a video quality optimisation model that is complete and consistent enough to drive reliable video encoder decisions. To use video quality measurement as a built-in feedback to the encoding decisions, there cannot be any room for error. After years of continuous testing with a group of ‘golden eyes’, we have identified the places where the video quality measurement model does not behave accurately during subjective testing. Whenever the model is found to be inaccurate enough to calculate the video quality of a specific scene, it is refined until the accurate measurement is reached. The result of this long and intensive effort is a highly intelligent and accurate algorithm that understands how to measure video quality independent of the scene type and the video details. Clearly, accuracy is key when it comes to objectively measuring video quality. When the algorithm described here is applied to video compression, content providers and Pay-TV operators can deliver consistent, high-quality video experiences at low bitrates. The workflow is described below: n Step 1: The video encoder looks to achieve best compression and maximise video quality based upon the available bandwidth. Applying accurate video quality measurement to compression

Figure 4. Barten model for contrast sensitivity as a function of spatial frequency

Figure 4 shows that HVS responds differently to each spatial frequency. The highest sensitivity is to contrast information at ~ 4 cpd. The diagram also depicts that sensitivity goes down to zero very quickly for high frequency. Sensitivity decreases for lower frequencies, but does not completely disappear. n Pixel Information: All of the pixel information in an image creates different types of masking on each of the pixels where distortion is measured. The plurality of certain information in an image reduces the sensitivity to each occurrence of the same information. Pixel information is characterised by its contrast, orientation and spatial frequency. Furthermore, the plurality of the total information in an image reduces the ability to detect the distortion in an image. It is a common practice of video encoders to increase the level of compression in images that contain more information. n Sequence Change: Video is a sequence of images. When the video is changing across the images that compose the video scene, the sensitivity to the distortion might increase or decrease depending upon the type and amount of changes. Taking into account the spatio-

Clearly, accuracy is key when it comes to objectively measuring video quality

47

Vol. 38 No. 4 - November 2016 Issue