Research On Deformed Object-of-interest Locate Algorithm Based On Extended Sparse And Part Contour

The detection and localization of Object-of-Interest (OOI) are always among the key issue of several fields, such as computer vision and pattern recognition etc. A lot of research work about non-deformable target localization algorithms has been carried out in the past few decades, but it's also difficult to detect deformable OOI. Based on some new thoughts and the summary and analysis of the relevant research, this paper puts forward three new OOI detection algorithms to solve some problelms during the process of object detection.Firstly, a localization algorithm of OOI based on sparse active contour model is presented in this paper, because the traditional object localization algorithm based on shape information is difficult to detect the object when the target is viewed from different angle or existing a little deformation. The sparse active contour model of OOI is learned by shared sketch algorithm, and Gabor basis that make up of the model can be adjusted by local perturbation to fit the image. Then the framework of sum maps and max maps in turn is computed to obtain the region which matching the active contour template with the highest scores is extracted in the testing image. Finally, extracted images are classified and identified according to visual cortex model. Experimental results demonstrate the effectiveness of the algorithm.Secondly, the sparse active contour model can be used to solve the problem of localization when the target existes a little deformation, but it is difficult to slove the problem when the target is large changed. This paper puts forward an object detection algorithm based on sparse active contour spread shape script model to slove this problem. Various deformable shape motifs are learned with sketch samples by the spread active contour model, and these motifs that make up of the shape script can be adjusted to fit test images, so it is robust when targets occurre large deformation. Then the matching of a shape script template to test images can be accomplished by a cortex-like structure of recursive sum-max maps, and localization of OOI is realized. The experiments show good performance of this method.Finally, a localization algorithm of OOI based on mixture model about HOG feature is proposed in this paper aiming at the problem that the traditional object localization algorithm based on HOG feature is always wrong when the target is slight deformed or interfered by noise. The classifiers LSVM are trained effectively by HOG feature of train images, and mixture models of OOI that include root models, part models and corresponding deformation models are learned at the same time. Then OOI is localized according to dynamic programming under which the matching region with deformation models in test images. The experiments show that this algorithm is efficient when the target is partly changed, occlusive or in complex background.