PROJECT TITLE :

Semantic Image Segmentation with Contextual Hierarchical Models

ABSTRACT:

Semantic segmentation is the matter of assigning an object label to every pixel. It unifies the image segmentation and object recognition issues. The importance of using contextual information in semantic segmentation frameworks has been widely realized in the field. We have a tendency to propose a contextual framework, known as contextual hierarchical model (CHM), that learns contextual info in an exceedingly hierarchical framework for semantic segmentation. At each level of the hierarchy, a classifier is trained based mostly on downsampled input images and outputs of previous levels. Our model then incorporates the resulting multi-resolution contextual data into a classifier to segment the input image at original resolution. This coaching strategy permits for optimization of a joint posterior probability at multiple resolutions through the hierarchy. Contextual hierarchical model is solely primarily based on the input image patches and will not build use of any fragments or shape examples. Hence, it's applicable to a variety of problems like object segmentation and edge detection. We tend to demonstrate that CHM performs at par with state-of-the-art on Stanford background and Weizmann horse datasets. It also outperforms state-of-the-art edge detection ways on NYU depth dataset and achieves state-of-the-art on Berkeley segmentation dataset (BSDS 500).