The Study Of Low-Rank Coding Algorithms For Robust Images Representation And Classification

Low-rank coding is an important research topic in the fields of computer vision,data mining,signal processing and pattern recognition.It has been successfully applied in image description,image de-noising,image retrieval and classification.In view of the existing low rank representation algorithm is sensitive to noise,distinguishing power is insufficient,and couldn't adaptively keep the neighbor information,this paper proposes more discriminative,stronger robustness and higher efficiency of low rank representation algorithm.The real data evaluate the validity of our algorithms in the application of image representation and classification.The innovation work of this paper mainly includes:(1)We technically propose a novel Similarity-Adaptive Latent Low-Rank Representation(SA-LatLRR)model.SA-LatLRR incorporates a reconstructive error minimization term over the coefficients and salient features,which can clearly preserve the neighborhood information among salient features adaptively.The shared coefficients could minimize the reconstruction errors over both original data and salient features at the same time,where the embedded salient features contain less noise and unfavorable features than the original data.To make the learnt salient features more informative and robust to noise,our SA-LatLRR imposes the sparse L2,l-norm and low-rank constraints on the projection jointly.(2)We propose a novel Robust Adaptive Low-rank and Sparse Representation(RALSR)framework for salient feature extraction of the high-dimensional data.Specifically,RALSR integrates the joint low-rank and sparse representation,adaptive neighborhood preserving graph weight learning and the robustness-promoting representation into a unified framework.For accurate similarity measure,RALSR computes the adaptive weights by minimizing the reconstruction error over the noise-removed data and salient features.RALSR can also ensure the learnt projection to preserve local neighborhood information of embedded features clearly and adaptively.making it powerful for the salient feature extraction.(3)We propose a Robust Adaptive Structure-constrained Low-Rank Coding(AS-LRC)mod-el.To recover the underlying subspaces accurately,our AS-LRC seamlessly integrates the adaptive weighting based block-diagonal structure-constrained low-rank representa?tion and the group sparse salient feature extraction into a unified model.To enforce the block-diagonal structures adaptive to different real datasets,AS-LRC clearly computes an auto-weighting matrix based on the locality-adaptive salient features and multiplies by the low-rank codes for minimization.This encourages the codes to be block-diagonal and can also avoid the tricky issue of choosing optimal neighborhood size or kermel width for the weight assignment.In addition,AS-LRC selects the L2,1-norm the pro-jection for extracting group sparse features which can make the learnt features robust to noise and outliers in samples,and can also make the feature coding process efficient.