Image Recognition Algorithm Based On Nearest Reduction And Dimension Reduction

In pace with the development of science and technology, dimension and the number of sample increase quickly. Then computer storage and computing seed face much challenge, and Information explosion leads to redundant information, in recent years, the ideas of samples reduction and dimension reduction have been researched as a popular topic. On the basis of analyzing the related domestic and international research results, this paper studies the sample reduction and dimension reduction.Firstly, because the number of sample is large and pseudo near neighbors disturb sample classify, We apply large margin nearest neighbor Classification based on sample reduction. Sample that far away from classification boundary has little effect on classification, So we use the chain of nearest neighbor eliminate the sample that is far away from classification boundary. And we encountered the pseudo nearest neighbors very often, we can learn the Transfer matrix through objective function of large margin nearest neighbor that makes the neighbors with same class label close while the neighbors with different class label far away.Secondly, Locality Preserving Projections need to choose the parameters, but the parameters are difficult to choose. And classification result is easily affected by noise. We use the geometrically motivated assumption that for each data point there exists a small neighborhood in which only the points that come from the same manifold lie approximately in a low-dimensional affine subspace. So we can make use of sparse optimize on affine space, though that we can get the nearest neighbors and the weights without setting parameter K.Finally, Image are presented by vectors, the dimension is very high, So can’t apply the classic classify methods. Inspired by singular value decomposition and non-negative matrix factorization, etc, So we learn the basis matrix by exploiting the intrinsic geometric structure of the data. By applying the spectral analysis on the nearest neighbor graph, we get the basis vectors, then we use lasso to learn the sparse representation with the learned basis for each image.