Semantics extraction from images
Knowledge-driven multimedia information extraction and ontology evolution
| Hi-index | 0.00 |
The traditional processing flow of segmentation followed by classification in computer vision assumes that the segmentation is able to successfully extract the object of interest. This is extremely difficult without any prior knowledge about the object that is being extracted from the scene. We propose a method of segmentation that uses the classification subsystem as an integral part of the segmentation, which will provide contextual information regarding the objects to be segmented. We note that traditional segmentation can then be viewed as a filter operating on the image independently of the classifier, much like the filter methods for feature selection. Our motivation for integrating segmentation and classification follows the wrapper methods of feature selection. In the wrapper methods for feature selection, the classifier is an integral part of the selection process and serves as the metric to decide the best feature set. In the same way, we wrap the segmentation and classification together, and use the classification accuracy as the metric to determine the best segmentation. We show the performance of wrapper-based segmentation on real-world and complex images of automotive vehicle occupants.