Bornomala: A Deep Learning-Based Bangla Image Captioning Technique

Image captioning, giving a description of an image is becoming popular nowadays. Bangla is spoken by almost 260 million people on the planet. However, a little effort or research has been done on it in image captioning area. In English, just the opposite, there are numerous automatic photo captioning systems. Our motive is to create a Bangla automatic image captioning system that can correctly describe any image in Bengali. In image captioning, it is important to acknowledge the important objects of the image and how they are related to an image. Moreover, significant portions of the image are observed at first and then a caption is generated according to them. It is also important to generate semantically and syntactically right sentences. In this paper, we will try to present an attention based deep Bangla image captioning technique. Here, CNN is utilized to find out the features from the picture and their connections. Then LSTM is utilized to create a text description of the input image using this object and attention value. The result showed that, our model can predict the description of any image in Bangla very significantly. We also reported the BLEU score is between 35–44 which indicates that the model can generate the description for any image very rightfully.