CNN视觉特征的图像检索

Image Retrieval Based on CNN Visual Features

卷积神经网络(CNN)是当前图像识别领域的研究热点,利用预训练的CNN网络提取的图像特征展示出了较强的图像识别能力.主要对比分析了传统视觉特征和CNN视觉特征在基于内容图像检索任务中的性能表现,并指出了一些可以值得深入研究的方向.在两个公开数据库(Pascal Sentence和Pascal VOC 2007)的实验尝试表明CNN视觉特征比传统的视觉特征更适用于图像检索.

Convolutional neural network （CNN） currently becomes research focus for image recognition. The visual features extracted from the pre-trained CNN demonstrate powerful ability for various recognition tasks. The performance of traditional visual features and CNN visual features for content-based image re- trieval was mainly compared. Experiments on the two public available datasets of Pascal Sentence and Pascal VOC 2007 show that, a sufficient performance of CNN visual features used in image retrieval when compared with traditional visual features.