Research On Feature Extraction Algorithm For Low-Quality Images

With the rapid development of computer vision and machine learning methods,image recognition related technologies are increasingly complex and diversified.However,the images collected in reality usually have problems such as noise interference and image corruption,which increases the difficulty of image feature extraction.Based on the phenomenon of noise interference,white block corruption and stripe corruption existing in the image,the feature extraction algorithm of low-quality images(noise image and damaged image)is deeply studied.The research work is as follows:Due to factors such as equipment temperature or transmission media,there may be redundant interference information in the image,resulting in blurred images,and it is difficult to obtain useful information.Aiming at this problem,an image feature extraction algorithm for image optimization and discriminant projection is proposed.Firstly,in order to overcome the limitation of dimension and poor robustness of linear discriminant analysis,L_2,1 norm is introduced as the metric criterion to redefine the within-class scatter matrix and between-class scatter matrix.At the same time,in order to eliminate the adverse effects of edge class samples,the between-class scatter matrix is weighted and smoothed;Then,the prior knowledge is used to describe the neighborhood relationship between the samples to build a more accurate neighborhood graph;Finally,the experimental results on several public image datasets show that compared with the comparison algorithm,the proposed algorithm has higher classification accuracy and better result.Aiming at the problem that the important information of the image may be lost due to factors such as equipment failure or man-made corruption,which makes the image difficult to distinguish,a dynamic low-rank preserving projection image feature extraction algorithm is proposed.Firstly,in order to solve the problem of dimensionality limitation,the projection matrix is decomposed to obtain flexible feature dimensions.At the same time,in order to reduce the computational burden of solving the kernel norm,the matrix fast three decomposition theory is used to decompose the low-rank matrix;Then,in order to solve the problem that the predefined dictionaries may not be representative,a method based on learning dictionaries is proposed;Secondly,in order to maintain the local structure and reduce the computational complexity,graph constraints are imposed on the data reconstruction errors;Finally,the experimental results on several public image datasets show that the proposed algorithm has better classification effect than the comparison algorithm.At the same time,compared with similar algorithms in the comparison algorithm,the proposed algorithm has lower time consumption.