Saliency analysis based on multi-scale wavelet decomposition

In this paper, an efficient saliency analysis method based on multi-scale wavelet analysis is proposed, which can be used for car detection in intelligent transportation applications. In our method, saliency regions are considered as abnormal parts in a normal background; the wavelet theory is then used to detect these abnormal parts. Compared with other wavelet-based methods, our method need not perform inverse wavelet transformation, which is time-consuming. Besides, the use of multi-scale wavelet analysis can eliminate the drawbacks of the traditional center-surround methods, which have difficulties in detecting salient regions far away from object boundaries. Furthermore, a saliency prior process is adopted in our method, which can enhance the saliency map. Experimental results show that our method can achieve excellent results in terms of receiver operating characteristic (ROC) curve, the area under the curve (AUC) score, and visual performance, as compared to other state-of-the-art methods.