IMAGE RETRIEVAL BASED ON MULTIPILE FEATURES AND LINEAR DISCRIMINANT ANALYSIS
The fully connected layer features of convolutional neural networks lack the description of the underlying information of image,result in that some samples cannot be retrieved successfully.Moreover,it's not efficient to retrieve directly due to the high dimension of the fully convolutional layer features.To solve this problem,we propose an image retrieval method by combining linear discriminant analysis and multiple features.Convolutional neural network was used to extract features from both convolutional layer and fully convolutional layer,and it merged HSV color features together.Dimension of the fused features was reduced by linear discriminant analysis.Multi-layer features could increase the image differentiation and improve the recognition accuracy.Comparison experiments show that this method makes improvement in precision and speed on the task of image retrieval.
Deep learningMultiple featuresLinear discriminant analysisImage retrieval
国家科技期刊平台
NSTL
万方数据
维普
