Image retrieval method based on deep learning semantic feature extraction and regularization softmax

Abstract

In content-based image retrieval (CBIR), an image retrieval method combining deep learning semantic feature extraction and regularization Softmax is proposed for the “semantic gap” between the underlying visual features and high-level semantic features. First, the deep Boltzmann machine (DBM) and the convolutional neural network (CNN) in the deep learning method are combined to construct a convolution depth Boltzmann machine (C-DBM), which enables it to extract High-order semantic features of images, and robust to image scaling, affine and other transformations. Then, the Dropout regularized Softmax classifier is used to classify the image features. Finally, the image is retrieved according to the sort output. The experimental results show that the proposed method can extract semantic features effectively and has high retrieval accuracy. The classification precision rate in STL-10 image data set reaches 60.3%.