基于Gabor滤波器和深度学习的图像检索方法

针对图像数据库日渐庞大的问题,研究了将特征提取与深度学习相结合进行图像检索的方法,提出了基于Gabor小波变换和受限玻尔兹曼机(RBM)的特征提取和降维模型.将整幅图像划分成局部图像块,利用Gabor滤波器组提取图像特征,通过RBM对特征进行学习和编码,从而实现图像特征的降维处理.采用基于深度信念网络(DBN)和Softmax分类器的图像检索算法,利用Corel图像库进行新方法的图像检索实验,并与其他两种方法进行比较.结果表明,本文方法在准确率、查全率和检索时间上均具有较好的性能,能得到更好的图像检索结果.

Image retrieval method based on Gabor filter and deep learning

To solve the problem that the image database is becoming larger, an image retrieval method combined with both feature extraction and deep learning was investigated, and a model for feature extraction and dimensionality reduction was proposed based on Gabor wavelet transformation and restricted Boltzmann machine(RBM). The whole image was divided into local image blocks, and a set of Gabor filters were used to extract the image features, and the image features were studied and encoded with RBM. Hence, the dimensionality reduction of image features could be achieved. An image retrieval algorithm based on both deep belief networks(DBN)and Softmax classifier was adopted. In addition, the Corel image database was used to perform the image retrieval test for the new method, and was compared with other two methods. The results show that the proposed method has better performance in precision rate, recall rate and retrieval time, and can obtain better image retrieval results.