Weakly Supervised Image Annotation and Segmentation

Abstract

The various aspects in the processing of an image include object recognition, object classification, image segmentation, and attribute learning, are closely related to each other. In this paper, we proposed a Bayesian Non-parametric (BN) approach to solve the complex visual tasks using the non-parametric property to regulate the model's constraint. A Chinese Restaurant Process Stacked with Weakly Supervised Markov Random Field (WS-MRF-CRP) is developed, which uses Markov Random Field (MRF) for low-level and Chinese Restaurant Process (CRP) for high-level. The proposed approach learns and incorporates association between various object and attribute classes. The input image is clustered into individual components using the MRF, and then the CRP is used for merging the components and generating the image-attribute association. Experiments performed on the Berkeley Segmentation dataset demonstrated that the proposed model performs better than other existing weakly supervised models. © 2021 IEEE.