Research On Image Recognition Based On Nonnegative Weighted Locally Linear KNN Model

In the field of computer vision,the image recognition technology plays a very important role in a variety of application scenarios.Therefore,many researchers have proposed many methods of image recognition,one type of the most successful methods is sparse representation(SR)methods,and the sparse representation has a good effect on the problem of image classification.Image recognition algorithm based on local linear KNN model(LLKNN)is robust and has good performances.It considered the construction,locality and sparsity simultaneously to learn sparse representation coefficients for image recognition.It adopts the linear combination of all the training samples to represent the test sample.The representation coefficients of LLKNN preserve the grouping property of the nearest neighbors,and the model establishes a relation of sparse coefficient and classification to propose the LLKNN based classifier(LLKNNC)and the local linear nearest mean classifier(LLNMC).In this paper,we mainly study the local linear KNN model thoroughly,and the corresponding improved algorithms are put forward to image recognition problems.Because the non-zero representation coefficients obtained by LLKNN model can be positive or negative sign.In many practical situations,negative representation coefficients are meaningless and unreasonable,in order to improve the performance of LLKNN model and remove the effect of negative representation coefficients.Therefore,based on LLKNN,we propose an image recognition algorithm named nonnegative locally linear KNN model(NLLKNN)by introducing non-negative constraints in our first research work.In our second research work,the correlations between test samples and training samples are regarded as constraints and propose a weighted locally linear KNN model(WLLKNN)for image recognition by adding a reasonable weight to representation coefficients of LLKNN.In our third research work research,we propose an image recognition algorithm,which is named the nonnegative weighted locally linear KNN model(NWLLKNN)by adding nonnegative restrictions and weights simultaneously.The research content of this article is summarized as follows:(1)In order to improve the classification performances of locally linear KNN model(LLKNN)for image recognition and make the representation coefficients more reasonable,based on the LLKNN algorithm,we introduce nonnegative constraints and propose a nonnegative locally linear KNN model(NLLKNN)for image recognition.In this paper,we not only give an iterative updating algorithm to solve the nonnegative sparse coefficient,but also give the proof of the convergence of the objective function.The experimental results show that,compared with the original LLKNN algorithm,the NLLKNN algorithm can obtain better performance and more robust.(2)In order to make the representation coefficients of LLKNN method sparser,considering the correlation between the test sample and the training samples,we introduce weighted representation coefficients to locally linear KNN model and propose a weighted locally linear KNN model for image recognition(WLLKNN).By the means of weighting on representation coefficients,the representation coefficients,corresponding to representation coefficients that are highly correlated with training samples,are increased.Otherwise,the representation coefficients will be compressed.WLLKNN algorithm not only gives five different weighting forms,but also gives a simple and effective iterative algorithm to solve the representation coefficient and the theoretical proof of the algorithm.Experiments on several different image database shows that the recognition performance of the WLLKNN algorithm is improved compared with the traditional LLKNN method.(3)Based on above,the nonnegative locally linear KNN model and the weighted locally linear KNN model can obtain better recognition results.Therefore,based on the locally linear KNN model,we combine the nonnegative constraints of representation coefficients and weighting the representation coefficients,and propose a nonnegative weighted locally linear KNN(NWLLKNN)for image recognition.In this paper,an iterative updating algorithm is given to solve the sparse representation coefficients of the NWLLKNN model,and the theoretical proof of the algorithm is given.The experimental results on multiple image databases verify the effectiveness of the NWLLKNN algorithm.