A saliency map generation method and an image processing system using the same are provided. The method includes the following steps. An original underwater image is received. A blurring processing is performed on the original underwater image to generate a defocus map. The defocus map is input to an auxiliary convolutional network model to obtain multiple feature maps of the defocus map. The original underwater image and the feature maps are input to a main convolutional network model to generate a saliency map of the original underwater image.