PROJECT TITLE :

Feature Map Quality Score Estimation Through Regression - 2018

ABSTRACT:

Understanding the visual quality of a feature map plays a important role in many active vision applications. Previous works mostly rely on object-level features, such as compactness, to estimate the standard score of a feature map. But, the compactness is leveraged on feature maps created by salient object detection techniques where the maps have a tendency to be compact. Hence, the compactness feature fails when the feature maps are blurry (e.g., fixation maps). During this Project, we tend to regard the process of estimating the standard score of feature maps, specifically fixation maps, as a regression drawback. When extracting many local, international, geometric, and positional characteristic features from a feature map, a model is learned employing a random forest regressor to estimate the quality score of any unseen feature map. Our model is specifically tailored to estimate the quality of three varieties of maps: bottom-up, target, and contextual feature maps. These maps are produced for a giant benchmark fixation data set of a lot of than 90zero challenging outdoor images. We demonstrate that our approach provides an accurate estimate of the standard of the abovementioned feature maps compared to the groundtruth knowledge. Yet, we have a tendency to show that our proposed approach is useful in feature map integration for predicting human fixation. Rather than naively integrating all three feature maps when predicting human fixation, our proposed approach dynamically selects the simplest feature map with the best estimated quality score on a personal image basis, thereby improving the fixation prediction accuracy.