Context-Based Approach For Multi-Captioning Object Annotation With Inference Knowledge Using Hidden Markov Model

Context-based inference offered a deeper understanding of the objects that exist in the image. It provides information about the object by considering importance, location, and its size. In order to identify the effective target objects based on context-aware captions, the Hidden Markov Model (HMM) is introduced in this work. This probabilistic model has the capability of linear categorization of objects within the image. Initially, the input image is divided into multiple segments for multi-caption object detection. After that, the objects in each segment are identified and represented in the states-of-state transition diagram. Finally, HMM identifies the context-aware linear sequence of objects based on transition and emission probabilities. The developed HMM inference model is checked for images with caption-annotated sequences that exist in the Flickr Image dataset. The results show that the proposed HMM offered appropriate annotations for the input images by considering the importance of the object, location, and size.