Fast and smart object proposals for object detection
Object localization plays an important role in object detection and classification. In the last years, several methods have shifted from sliding windows techniques to object proposals techniques. The latter produces a small set of windows submitted to an object classifier to reduce the computational time. In this paper, we propose a fast unsupervised method that combines the edge feature and saliency map to generate less than hundred bounding boxes from the processed image. Our approach exploits a number of rules based on edges information plus saliency regions to decide if an object is present in a window. We have carried out several experiments to validate our approach on ImageNet dataset and obtained very promising results.
Object proposals, Bounding box, Object detection, Saliency map, Sliding window