Region extraction in vector images

摘要: A semantic object tracking method tracks general objects with multiple non-rigid motion, disconnected components and colors throughout a vector image sequence. The accurately these by spatially segmenting regions from current frame then classifying as to which they originated in the previous frame. To classify each region, perform region based motion estimation between segmented computed position of predicted classifies being part on contains most overlapping points region. Using this method, is tracked one frame, no gaps or overlaps. propagates few errors because it projects into where boundaries are previously rather than trying project adjust boundary object's unknown.