Computer vision algorithms have complex set of parameters that significantly influence resulting quality. On the one hand, such a big amount of parameters that can be varied results in algorithm ability to fit for par...
详细信息
ISBN:
(纸本)9781665404761
Computer vision algorithms have complex set of parameters that significantly influence resulting quality. On the one hand, such a big amount of parameters that can be varied results in algorithm ability to fit for particular usage circumstances. On the other hand, fine-tuning of even several parameters may require a lot of human resources unless this task is automated. An approach to such automatization is provided considering a machine learning problem with several peculiarities. First, some parameters are restricted to have integer values. Second, multiple metrics describing algorithm quality are restricted to have integer values as well. Third, there are no analytical formulae describing dependences between parameters and metrics values. Mentioned features make gradient-descent-like methods widely used for neural networks training inapplicable for solving that optimization problem. As result, automatically tuned parameters provides better quality than manually selected ones.
Many distributed multimedia applications rely on video analysis algorithms for automated video and image processing. Little is known, however, about the minimum video quality required to ensure an accurate performance...
详细信息
Many distributed multimedia applications rely on video analysis algorithms for automated video and image processing. Little is known, however, about the minimum video quality required to ensure an accurate performance of these algorithms. In an attempt to understand these requirements, we focus on a set of commonly used face analysisalgorithms. Using standard datasets and live videos, we conducted experiments demonstrating that the algorithms show almost no decrease in accuracy until the input video is reduced to a certain critical quality, which amounts to significantly lower bitrate compared to the quality commonly acceptable for human vision. Since computer vision percepts video differently than human vision, existing video quality metrics, designed for human perception, cannot be used to reason about the effects of video quality reduction on accuracy of video analysis algorithms. We therefore investigate two alternate video quality metrics, blockiness and mutual information, and show how they can be used to estimate the critical video qualities for face analysisalgorithms.
暂无评论