Abstract:

It is difficult to accurately describe the full contour of targets with a complex structure in the cluttered scene, because of the complexity of structures, inter class differences, variation of conditional of images, background interference and so on. In order to overcome the shortages, a novel hierarchical model is proposed: in the lower layer, the conditional random fields model is used to fuse sorts of features to get the candidate regions of targets, which is helpful for high level processing; in the higher one, the distribution of shape of targets can be computed through statistic of the characteristics of parts of the model. Competitive results demonstrate the precision, robustness, and effectiveness of the proposed method.