Boosted Detection of Objects and Attributes

We present a new framework for detection of object and attributes in images based on boosted combination of primitive classifiers. The framework directly minimizes the detection error by learning a set of simple, computationally efficient threshold-based detectors. We apply this framework to segmentation of human skin and detection of faces in images. We show that despite its simplicity the method performs on par with more complex traditional models in detection accuracy while outperforming them in scalability. This can be especially beneficial in applications of the framework to real-time detection tasks.