Visual saliency plays an important role in biological vision. This paper proposes a novel feature-based approach of visual saliency detection for natural color images. The saliency for each pixel is determined by two features: the structure information and the color information, which represent local information of pixels. The structure information is represented by local steering kernel and the color information is represented by color values of local region around each pixel. We compare our method with the current state-of-the-art models on natural color images. Experiments on human visual fixation data and psychological patterns demonstrate that our method performs competitively for visual saliency detection task.