Delegating Custom Object Detection Tasks to a Universal Classification System

In this paper, the concept of multipurpose object detection system, recently introduced in [1], is discussed. Fig.1 below illustrates the innovative and the business aspects of this method: transform a classifier into an object detector/locator via an image grid. Classification techniques allow the application of powerful machine learning methods detailed below. A custom object detection system only needs to analyze the local classification results on the image grid. In this way, creation of a custom system is facilitated by first applying the universal approach and then doing a custom analysis of its results.