In order to solve the problem of poor robustness of Kernelized Correlation Filter (KCF) in complex scenes, a new object tracking algorithm based on Self-Adaptive Compound Kernel (SACK) was proposed. The tracking task was decomposed into two independent subtasks:position tracking and scale tracking. Firstly, the risk objective function of SACK weight was constructed by using the self-adaptive compound of linear kernel and Gaussian kernel as the kernel tracking filter. The weights of linear kernel and Gaussian kernel were adjusted adaptively by the constructed function according to the response values of kernels, which not only considered the minimum empirical risk function of different kernel response outputs, but also considered the risk function of maximum response value, and had the advantages of local kernel and global kernel. Then, the exact position of object was obtained according to the output response of the SACK filter, and the adaptive update rate based on the maximum response value of object was designed to adaptively update the position tracking filter. Finally, the scale tracker was used to estimate the object scale. The experimental results show that, the success rate and distance precision of the proposed algorithm are optimal on OTB-50 database, which is 6.8 percentage points and 4.1 percentage points higher than those of KCF algorithm respectively, 2 percentage points and 3.2 percentage points higher than those of Bidirectional Scale Estimation Tracker (BSET) algorithm respectively. The proposed algorithm has strong adaptability to complex scenes such as deformation and occlusion.

To overcome the limitation of the assumptions of topic model-word independence and topic independence, a topic model which inosculated the spatial relationship of visual words was proposed, namely Markov Topic Random Field (MTRF). In addition, it was discussed that the "topic" of topic model represented the part of object in image processing. There is a high probability of the neighbor visual words generated from the same topic, and whether the visual words were generated from the same topic determined the topic was generated from Markov Random Field (MRF) or multinomial distribution of topic model. Meanwhile, both theoretical analysis and experimental results prove that "topic" of topic model appeared as mid-level feature to represent the parts of objects rather than the instances of objects. In experiments of image classification, the average accuracy of MTRF was 3.91% higher than that of Latent Dirichlet Allocation (LDA) on Caltech101 dataset, and the mean Average Precision (mAP) of MTRF was 2.03% higher than that of LDA on VOC2007 dataset. Furthermore, MTRF assigned topics to visual words more accurately and got the mid-level features which represented the parts of objects more effectively than LDA. The experimental results show that MTRF makes use of the spatial information effectively and improves the accuracy of the model.