A Novel Bi-LSTM Based Automatic Image Description Generation

Image captioning is the process of creating a textual description of an image. Due to its importance in various fields, it has emerged as the latest and hot research problem. It uses Computer Vision techniques to process an image and Natural Language Processing to generate the caption. Our proposed approach uses the Bi-LSTM (Bi directional Long Short Term Memory) approach to generate the image description. We also propose the Novel Moth Flame Optimization (NMFO). This model uses the correlation-based logarithmic spiral update. The novel proposed model is demonstrated on standard datasets like Flicker 8k, Flicker 30k, and MSCOCO datasets using standard metrics likeBLEU, CIDEr. Performance of various metrics on various datasets shows that our novel Bi-LSTM approach gives better performance when compared to our traditional approaches.