Wavelet moment and improved adaboost application to vehicle-logo location

In this paper, we firstly describe improved Adaboost algorithm, and then introduce wavelet moment, which has rotation, shift, scale moment invariants and multi-resolution characteristic, and which can not only extract image local feature, but also can extract global feature. so it is more stronger to oppose noise. This paper proposes dynamic pyramid changing idea of detector scale. A cascade classifier was trained based on wavelet moment and improved Adaboost to detect Vehicle-logo. The experiment proved that it was real-time and had quite a high rate of detection.