Real-time computation of moment invariants combined with contrast stretching

There is a large body of evidence that demonstrates the practicable use of moment invariants in real-time computer vision applications. Object detection, recognition and tracking are some of them, to name a few. However, the efficacy of moment invariants is highly susceptible to varying illumination conditions, which is inherent in real-world applications. Contrast stretching alleviates the problem, but performing contrast stretching prior to the calculation of moment invariants is computationally intensive and not suitable for real-time use. We address this problem by proposing an efficient real-time method that integrates the calculation of moment invariants up to the 4th order with a contrast stretching operation (all in one go), by utilising Summed-Area Tables (SATs). The method is limited to general contrast stretching, not necessarily covering very distinct illumination sources from different directions; that is, illumination conditions that create extra strong edges. We test the proposed method with a popular benchmarking image database (Amsterdam Library of Object Images) that is publicly available. Such images were acquired in a controlled environment, demonstrating varying lighting conditions. We show empirically that the method works in real-time and accurate enough for practical object detection applications.