Image Captioning: From Encoder-Decoder to Reinforcement Learning

With the rise of artificial intelligence technology, both image feature extraction technology and text automatic generation technology have made great progress, and the image description generation technology that combines the two has also been attached burgeoning importance by academic fields. It enables computer to have the ability to catch features of a picture and generate a sentence, which is another new type of task after relaying image recognition, image segmentation, and target tracking in image processing. This paper provides a few popular datasets and methods of feature extraction. Moreover, some improving methods, based on Encoder-Decoder, Attention Mechanism, and Reinforcement Learning, are discussed. In the end, this paper provides some applications as examples accomplished by image caption.