Research And Application On Object Extract Algorithm Based On Target Spatio-Temporal Analysis

Object extraction belongs to medium-vision areas. By detecting significant results with independent in the input sequences, which can be provided as better input in high-level vision tasks. Also it is a very important research topics. Based on the temporal and spatial analysis of image sequences, considering the motion, color, texture, and other temporal characteristics, we proposed a significant research target extraction algorithm, and apply it to detect, identify and target areas such as video segmentation.In this paper, an image sequence as input, the target proposed extraction algorithm based on spatial and temporal analysis, the main work is as follows:1、Since the normalization similarity problem have much relationship with the results of clustering, the long trajectory of space-time structure characteristics were studied in this paper, we proposed a significant target extraction algorithm based on an adaptive scaling factor. Compared with Brox algorithm results, the number of dead pixels in the foreground and background are reduced a lot;2、For the problem that a lot of long-points track in Brox algorithm are not able to reflect significant target in the image, and too many long point trajectory also have a significant impact on the efficiency of solving similarity and clustering. In this paper, the use of long point edge features to optimize trajectory, based on the complexity of the issues raised on the edge of a significant length of track target contour grouping algorithm, which not only can reduce the calculation of similarity and clustering, but also the resulting contour grouping results better reflect the image of the target significantly;3、Finally, the definition of similarity for the long point trajectory is difficult and the number of point trajectory too many led to similarity calculation and clustering complex issues. In response to these problems, the paper focus on the edge extraction processing directly to obtain multi-feature similarity edge segments. We proposed a novel edge contour grouping algorithm based on multi-feature of edge. The Algorithm can not only solve the above problems, but also extract more precise results.