A probabilistic and statistical method for analysis of MPEG-encoded video.
|Title:||A probabilistic and statistical method for analysis of MPEG-encoded video.|
|Abstract:||To achieve high rates of compression, the MPEG video compression standard provides methods for reducing both spatial and temporal redundancies in video data. The key step in lowering spatial redundancies is the application of the two-dimensional Discrete Cosine Transformation, while calculation of motion vectors is vital for temporal dependence reduction. Both the DCT coefficients and motion vectors can be used to provide information about the video data itself, without actually viewing or even completely decompressing the original data. After discussing different models for background noise, statistical hypothesis tests are developed to detect either the presence or motion of an object in a small area of one picture in the video sequence using either the DCT or motion vector data. The hypothesis tests are then combined into a unified cumulative sum procedure, based on the p-values of the hypothesis tests, which will signal an alarm at the point in time when an object appears in the video under investigation.|
|Collection||Thèses, 1910 - 2010 // Theses, 1910 - 2010|