Research On Image Recognition Based On Subspace Reconstruction

Image recognition is an important direction in many research fields of pattern recognition and computer vision.Image recognition includes different recognition tasks such as face recognition,target recognition and motion recognition.Image recognition technology has played a very critical role in real-time applications in various fields of life,such as various reconnaissance and security systems in the military and public security criminal investigations,and smart household appliances used for home services in life.Image recognition technology can be applied to various scenes,but affected by image quality and scene complexity,the current image recognition algorithm is mainly to improve the accuracy of image recognition for specific scenes.In addition,the accuracy of image recognition is also affected by preprocessing,feature extraction,image augmentation,and classification algorithms.The main content of the image recognition algorithm discussed in this paper is the algorithm based on sparse coding and deep sparse coding based on subspace reconstruction.The inquiry details and new ideas involved in the article include the following two directions:First,to deal with the image recognition problem of video sequences,sparse coding based on classification can obtain the sparse solution of the target,making the resulting classification results more robust.This paper combines the advantages of bilinear regression method and sparse coding method,trying to find the "approximate solution subspace" of all categories,reducing the interference of similar categories,so as to improve the image retrieval method more suitable for different classification tasks.The experiments used in this paper are the three data sets Caltech101,You Tobe and LFW,which are widely used in image classification.Experiments have proved that the algorithm proposed for subspace reconstruction is feasible.Second,the method of manually extracting image features is a method commonly used by traditional image recognition algorithms in extracting image features.The method of manually extracting images works well under certain conditions.However,when the image object under study changes,the method of manually extracting image features cannot flexibly extract the effective features of images in various environments.At present,a very popular method for extracting image features is to use neural networks.In practical applications,the related algorithms for extracting image features using neural networks are also the most extensive.This paper designs an image classification and recognition algorithm based on deep sparse coding.The main framework used in this algorithm is the structure of sparse coding connecting neural network Encoder and Decoder.This paper also designed an auxiliary classification algorithm.The purpose of the classification-assisted algorithm is to introduce the idea of subspace.Its role is to guide the neural network to extract the features of images of the same category in the training stage to be more similar,thereby improving the accuracy of image recognition.This algorithm made an experimental comparison on the two data sets of USPS and UMD,and achieved a good recognition effect.