An Image-Text Association Matching Algorithm Based on CRNN

The text in the pictures accompanying Internet rumors always doesn't match the rumors themselves. In order to improve the efficiency of Internet rumor detection, we propose an image-text association matching algorithm based on CRNN. The algorithm first determines the image suffix format and URL, and then uses a model that uses the fusion of CNN, RNN and CTC to identify the image. Among them, CNN performs feature extraction on input fake news images. RNN trains feature sequences and predicts label distributions, CTC converts all obtained label distributions, and finally get the text in the image. Then use the formatting algorithm to get the matching degree between the image and the text to associate both of them. Experiments show that our algorithm can achieve an accuracy of 94 % in matching pictures and press releases, which can effectively solve the problem of online rumor detection. Experiments show that our algorithm can achieve an accuracy of 94 % in matching pictures and press releases, which can effectively solve the Internet rumor detection problem.