EnMS: Early non-Maxima Suppression

EnMS: Early non-Maxima Suppression

en

Detection of objects in images using statistical classifiers is a well studied and documented technique.  Different applications of such detectors often require selection of the image position with the highest response of the detector -- they perform non-maxima suppression.  This article introduces the concept of Early non-Maxima Suppression, which aims to reduce necessary computations by making the non-Maxima Suppression decision early based on incomplete information provided by a partially evaluated classifier. We show that the error of one such speculative decision with respect to a decision made based on response of the complete classifier can be estimated by collecting statistics on unlabeled data.  The article then considers a sequential strategy of multiple early non-Maxima suppression tests which follows the structure of soft-cascade detectors commonly used for object detection. We also show that an optimal (fastest for requested error rate) suppression strategy can be created by a novel variant of Wald's sequential probability ratio test (SPRT) which we call the Conditioned SPRT, CSPRT.  Experimental results show that the Early non-Maxima Suppression significantly reduces amount of computation in the case of object localization while the error rates are limited to low predefined values. The proposed approach notably outperforms the state-of-the-art detectors based on WaldBoost. The potential applications of the early non-Maxima suppression approach are not limited to object localization and could be applied wherever the goal is to find the strongest response of a classifier among a set of classified samples.

Detection of objects in images using statistical classifiers is a well studied and documented technique.  Different applications of such detectors often require selection of the image position with the highest response of the detector -- they perform non-maxima suppression.  This article introduces the concept of Early non-Maxima Suppression, which aims to reduce necessary computations by making the non-Maxima Suppression decision early based on incomplete information provided by a partially evaluated classifier. We show that the error of one such speculative decision with respect to a decision made based on response of the complete classifier can be estimated by collecting statistics on unlabeled data.  The article then considers a sequential strategy of multiple early non-Maxima suppression tests which follows the structure of soft-cascade detectors commonly used for object detection. We also show that an optimal (fastest for requested error rate) suppression strategy can be created by a novel variant of Wald's sequential probability ratio test (SPRT) which we call the Conditioned SPRT, CSPRT.  Experimental results show that the Early non-Maxima Suppression significantly reduces amount of computation in the case of object localization while the error rates are limited to low predefined values. The proposed approach notably outperforms the state-of-the-art detectors based on WaldBoost. The potential applications of the early non-Maxima suppression approach are not limited to object localization and could be applied wherever the goal is to find the strongest response of a classifier among a set of classified samples.