Object Tracking Algorithm Research Using Contourlet Transform

The reason why the technology of video object tracking having been attached much attention is that it used in civil and military fields widely, such as video surveillance, military guidance, intelligent control and so on, for which, the research for video object tracking is becoming more and more important.Most of algorithms of video object tracking indicate the object using color feature, the performance of tracking video object will be affected by similar color object or illumination mutations. For which, it is necessary to use other features to indicate the object. Contourlet transform is a method of "sparse" image representation and it can describe more information of an image, such as edge profile, the texture feature that extracted from the transform can resist the interference of illumination mutations and similar color object. After Contourlet transform for an image, several sub-band images can be obtained and they can build Contourlet histogram which can be used as the object template, statistical feature of Contourlet coefficients can be used as the feature of object region. In this paper, the work for two representations of object area is as follows:(1) Contourlet histogram is used as an iterative parameter of mean shift algorithm. In order to solve the problem that the mean shift algorithm can’t adjust the size of kernel bandwidth, an item of update scale is added to the mean shift algorithm using the relationship between the information entropy of search object area and the size of object. A variable kernel bandwidth tracking algorithm is proposed based on Contoutlet histogram and information entropy for tracking the object whose size change. The experimental results show that the proposed tracking algorithm can not only overcome the interference of similar color non-object in the scene effectively, but also adapt to changes of size of the object steadily.(2) The mean and variance of coefficients of Contourlet high-frequency sub-band images can be regarded as high-frequency sub-band characteristics, characteristic parameters of Grey-Level Co-occurence Matrix extracted from low-frequency sub-band are regarded as low-frequency sub-band characteristics. These characteristics can be combined to form the feature vector of object. In order to solve the problem that most search algorithms can’t locate the object accurately, Genetic Algorithm is introduced for selecting candidate objects and matching with object template in this paper, then the best loaction of object can be obtained, as a result, a tracking algorithm is proposed based on Contourlet texture feature and Genetic Algorithm. The experimental results show that the proposed algorithm can track the object accurately in the search region and has a good robustness compared with other algorithms.