Image Feature Extraction And Recognition Method Based On LPPNet

With the development of computer technology,image recognition technology has been more and more widely used,it has become a research hotspot in the field of pattern recognition,and feature extraction is a key issue in image recognition.Researchers have proposed many feature extraction methods for image recognition,although most of the existing feature extraction methods have achieved promising performance in image recognition,they do not perform well when the training samples are corrupted with relatively large noise.In order to address this issue,this thesis deeply researches the image feature extraction and recognition methods in deep learning framework.The main works are as follows:1)An LPPNet learning network structure is proposed for image feature extraction and recognition.Specifically,the network uses LPP to learn filters,and then binary hashing and block histograms are used for indexing and pooling.Since LPP takes the local geometric structure and class information of the data into a account simultaneously,the learned filterscan extract more discriminative features that are suitable for image classification and recognition tasks.In addition,according to the corresponding Eigenvalues,this thesis designs a set of weights for the learned filters ofeach layer to obtain the weighted filters.As a result,the features extracted by the weighted filters have more discriminative ability.Finally,the nearest neighbor classifier based on chi-squared distance measure is used for image classification.The experimental results on Yale B,CMU PIE and FERET datasets show that,compared with the existing feature extraction methods,the proposed LPPNet is more robust and achieves better performance.2)According to the proposed LPPNet learning network,an image recognition system is designed.The system can not only use the input training sample images to train the LPPNet learning network,but also use the trained LPPNet learning network to treat the recognition images for extraction features and recognition.