Research On Object Recognition Based On Manifold Learning And Derived Kernel Model

With the rapid development of computer vision and pattern recognition technology, ob-ject recognition technology has been widely used in the military, aerospace, astronomical observation, intelligent video surveillance, etc. It has become a hot topic in the fields of com-puter vision and pattern recognition. This research has important theoretical and practical application values. In this paper, we investigate this problem by using manifold learning and hierarchical learning. The main contents could be summarized as follows.(1) This paper presents an algorithm for recognizing dim moving target in infrared image sequence. This algorithm, which is designed based on the tensor locality preserving projection (TLPP), accepts tensors as inputs. Not only does the proposed method inherit the attractive characteristics of the TLPP in terms of exploiting the intrinsic manifold structure, but also reduces both space complexity and time complexity. Experimental results on two IR image sequences demonstrate the effectiveness of the proposed method.(2) A new template selection method for Derived Kernel (DK) is proposed in this paper. The templates involved in the construction of the DK play an important role. Effective tem-plate sets can lead to better similarity measure. In our method, the redundancy is reduced and the label information of the training images is used. In this way, the proposed method can obtain compact template sets with better discrimination ability. Experiments on four standard databases show that the DK based on the proposed method achieves high accuracy with low computational complexity.(3) A hierarchical feature extraction method based on DK is given in this paper. The proposed method extracts an effective feature called Local Neural Response (LNR) by alter-nating between local coding and maximum pooling operations. The local coding can extract the local salient feature of image. The maximum pooling operation builds the translation in-variance into the model. We also show that other invariant properties, such as rotation and scaling, can be induced by the proposed model under quite mild assumptions. Moreover, a template selection algorithm is presented in order to reduce the computational complexity and improve the discrimination ability of LNR. Experimental results show that our method is robust to local distortion and clutter compared with state-of-the-art algorithms.(4) A new method for general object recognition was proposed in this paper. Firstly, a new image representation called Invariant Descriptor-based Local Neural Response (IDLNR) is obtained by a hierarchical learning method. The proposed method have the property that it excels at discriminating between objects from different categories while simultaneously being invariant to imaging variations as well as intra-class variability. Then the obtained image representation is passed to a linear classifier which is suitable for large databases for objects recognition. We show that the proposed method exhibits excellent recognition performance and outperforms several state-of-the-art methods on challenging databases. We demonstrate that our general object recognition method is robust to image variations and has a low sample complexity. Furthermore, the theoretical analysis of the invariance of IDLNR is given. It is helpful for understanding the characteristic of the proposed method.