Content-based attention ranking using visual and contextual attention model for baseball videos
IEEE Transactions on Multimedia - Special issue on integration of context and content
| Hi-index | 0.43 |
A very hardware and computational efficient method for frame motion characterization is presented. The method replaces the time consuming calculation of two-dimensional m×n picture elements with that of two one-dimensional vectors. This is made possible by mathematically operating the luminance values of vertical and horizontal lines as the characteristic values of x and y direction respectively; either by taking simply the weighted average values, or taking only those of the above-threshold values. By comparing these characteristic values taken from two frames in a sequence of moving pictures, one can extract the components of frame changes quantitatively into three factors; the factor caused by the motion of objects, or the factors caused by the panning, or zooming of the camera. The efficiency of this algorithm is examined with some practical examples, and application possibilities in an automatic motion tracking system and image stabilizer are discussed