Arabic Image Captioning: The Effect of Text Pre-processing on the Attention Weights and the BLEU-N Scores

Image captioning using deep neural networks has recently gained increasing attention, mostly for English langue, with only few studies in other languages. Good image model is required to automatically generate sensible, syntactically and semantically correct captions, which turn requires good models both computer vision natural language processing. The process more challenging case of data scarcity, languages complex morphological structures like the Arabic language. This was reason why limited number have been published captioning, compared those In this paper, an efficient learning proposed. addition, effect different text pre-processing methods on obtained BLEU-N scores quality generated images, as well attention mechanism behavior were investigated. Furthermore, “THUMB” framework assess captions used -for first time- captions’ evaluation. As shown results, a BLEU-4 score 27.12, achieved, highest results so far, captioning. best THUMB obtained, previously common images.