Image Tag Completion by Applying SPFCM Clustering on the Features Learned by Deep Convolutional Neural Networks

Image tag completion is a process that aims to simultaneously enrich the missing tags and remove noisy tags. many of the images have vague, incomplete and irrelevant tags. These untrusted tags, reduce the accuracy of image retrieval. Hence, in recent years, many tag completion algorithms have been proposed in order to access to the tags associated with the content of images. Due to the effectiveness of deep learning in many research fields, in this paper a deep convolutional neural network has been used to extract suitable visual and semantic features of images. Also, considering the challenges involved in loading a large-scale image databases in memory, a Single Pass Fuzzy C-Means clustering algorithm is used in order to compute visually similar images and refining the image’s tags according to similar samples. The results show the effectiveness of proposed approach in images tag completion.