Based On Feature Description Of Image Scene Classification Algorithm

Scene classification is designed to automatically mark the massive image data into different categories by computer,so as to realize the automatic classification of the images.In recent years,The scene classification methods and technical types of endless,more and more widely used.In addition,salient region detection has been successfully applied to the image classification and recognition,by more and more researchers attention.This paper introduced Salient region detection,feature extraction and feature coding of scene image.Firstly,we introduce the coding method of constructing geometric phrases and geometric phrases pooling to establish a middle image representation structure for images.Compared with the spatial pyramid method,the superpixel grid segmentation is divided along the edge of the target,and the same feature is basically located in the same grid block,as far as possible to ensure the integrity of the target.Through the experiment,this method can establish the correlation between the local features and improve the classification accuracy.Secondly,we propose a hybrid feature scene classification algorithm based on the edge image prominence,and combine the original feature and the feature features of the edge graph to get the final scene image features.In order to better extract the significant area,we use the image of the upper boundary as a background a priori,with the help of Markov absorption probability calculation of the initial significant map,by arranging the relationship between each of the super-pixel nodes in the image and the previous point of the image,a better graph is obtained,and the last significant graph is generated by three optimization techniques.Experiments show that the method can extract the richer features and improve the classification accuracy.Thirdly,the feature extraction and classification of different scene image databases are carried out by convolution neural network.Through the convolution neural network,the image database automatic learning feature,feature extraction and classifier are combined to optimize,Experiments show that the convolution neural network has achieved a good classification accuracy,so that the image classification task is greatly simplified.