Robust Dominant Motion Estimation Using MPEG Information in Sport Sequences

Compressed domain global motion estimation using the Helmholtz Tradeoff Estimator

“…In the first algorithm that used the Helmholtz Tradeoff Estimator for GME [7], the homography parameters were calculated by first estimating lines out of motion vectors and then calculating perspective transformations out of these lines. A line l i in implicit representation a i x + b i y + c i = 0 can also be described perspectively by a vector…”

“…To grant better comparability, PSNR values for error frames without motion compensation are also provided. For GME as preprocessing step for motion compensation we used the original algorithm [7], a modified version using eight lines, our new method and a very accurate pixel-based method using gradient descent [1]. As can be seen, our new method performs best among the selected block-based methods in every shown case concerning the accuracy of frame registration.…”

“…Thus, a higher order motion model with precision comparable to results of more complex algorithms can be estimated. One approach described in [7] uses the Helmholtz Tradeoff Estimator (HTE) on lines placed on motion vector positions as a first step for GME. Nevertheless, this two-step method leads to misestimations in presence of larger foreground objects as it takes only few selected motion vectors out of a complete motion vector field.…”

“…Section 3 outlines an optimization step for the HTE using differently complex motion models to reduce calculation load for GME. For comparison, the first GME method with HTE as shown in [7] is described shortly in Section 4. Experimental evaluation of the described methods and a comparison to a pixel-based approach concerning background PSNR curves and mean values can be found in Section 5.…”

Representative Frame Decoration using unsharp filter in video summarization

“…These features are used as powerful tool in video summarization, indexing, and retrieval. Key frames can be extracted by using color information [2][3][4][5][6], motion based information [7][8], audio information [9-12], textual features [13][14][15], and spatial-temporal information [16][17]. Audio information is often utilized together with image features.…”