一种基于视觉词袋模型的图像检索方法

AN IMAGE RETRIEVAL METHOD BASED ON BAG OF VISUAL WORDS MODEL

为了提高图像检索的效率,提出一种基于视觉词袋模型的图像检索方法。一方面在图像局部特征提取算法中,使用添加渐变信息的盒子滤波器构造尺度空间,以保留图像更多的细节信息,另一方面在特征表达时仅计算一次特征点圆形邻域内的Haar小波响应,避免了Haar小波响应的重复计算,并在保证描述子旋转不变性的同时做降维处理。同时,以改进k-means对特征库聚类构建加权的视觉词典,基于概率计算的方式选取k-means初始聚类中心,降低了传统k-means聚类效果对初始聚类中心选择的敏感性。实验结果表明该方法比传统方法具有更高的效率,特征提取速度提高48%左右,查准率提高2%以上。

In order to improve the efficiency of image retrieval, an image retrieval method based on BoVW （Bag of Visual Words） model is proposed. On the one hand, in image local feature extraction algorithm, we use box filter with gradient information to form scale space, to retain more image details information. On the other hand, only the Haar wavelet response in the circular neighborhood of the feature point is calculated in the feature expression, which avoids the repeated calculation of the Haar wavelet response and reduces the dimension while guarantee rotational invariance. At the same time, using improved k-means clustering method to construct a weighted visual dictionary, the k-means initial clustering center is selected based on probabilistic calculation method, which reduces the sensitivity of the traditional k-means clustering to the initial clustering center selection. The experimental results show that the proposed method is more efficient than the traditional method, feature extraction speed is increased by about 48% and the precision is improved by more than 2%.