Image saliency detection using red-black wavelet

The spectral residual method is well-known in image saliency detection for its simplicity and low computational cost, although the saliency map of traditional Fourier method is not fine-scaled and cannot offer an ideal detection result. To deal with this problem, we propose a frequency analysis method using red-black wavelet transform to obtain saliency map. The level of wavelet decomposition is determined according to the size of the image. After that the red-black wavelet forward transform is performed on both the original image and its Gaussian smoothing version. The difference between the two decomposition results is inverse transformed in order to obtain the saliency map. The experimental results demonstrate that our method performs a better result in saliency detection.