Research On Image Classification Based On SPM Model

At present,information generated by rapidly-developing artificial intelligence and the Internet includes image and video data.Because of the widespread use of multimedia technologies such as mobile phones and digital cameras,the growth rate of image and video data is too fast.Accurately searching for images required by people in the data becomes a focus of attention in the image processing field.Currently,image processing methods are broadly classified into image classification based on SPM(Spatial Pyramid Modal)model and image classification based on deep learning.In recent years,many researchers have conducted extensive research in the field of deep learning,but on the SPM model,research is rare.First of all,the dissertation describes the background,development and significance of image classification.The five parts of the framework of the spatial pyramid study the involved technologies,including pyramid partitioning methods,image feature selection,and so on.Secondly,the dissertation mainly studies the classification accuracy and classification time complexity of image classification,and improves the visual dictionary and feature encoding algorithm.In the visual dictionary construction stage,an efficient visual dictionary was obtained through the combination of K-means algorithm and hierarchical clustering algorithm.The K-means algorithm was used to perform preliminary clustering of data samples to obtain a rough division,and then used agglomerative hierarchical clustering.Accurate clustering was performed and the attribute weighting of information entropy was introduced in the algorithm.At the feature coding stage,the sparse coding,locally constrained coding and non-negative constrained coding were compared and studied,and improved on the basis of non-negatively constrained coding.The similarity weighting was applied to the coding and it was added to the coding with higher similarity.The weights make the description of image information more complete through this method,thereby improving the accuracy of image classification.Finally,the experiments and analysis are carried out.The experimental results show that K-means and hierarchical clustering combined algorithm and similarity-weighted non-negative local constraint coding algorithm can effectively improve the classification accuracy of the spatial pyramid model.