基于DCNN分类的图像相关度度量 postprint

Abstract: When measuring the similarity between images, the content of the physical features (Color Layout Descriptor, Gray Histogram Descriptor, etc.) may not be very comprehensive, so it is necessary to refer to the semantic information contained in image vision to measure the relativity between images. In this paper, we propose a method based on Deep Convolutional Neural Networks classification model to measure image correlation. The model is used to bind the semantic label from WordNet, and the label is filter and expand according to WordNet structure, and the concept set is used to calculate image relativity. Compared with the manually determined sample data, the peak value of Pearson correlation coefficient can reach 0.73, which proves that this method has a certain effect in the measurement of image correlation.