A superpixel-based saliency model for robust autofocus in low contrast images

Due to low brightness, the performance of autofocus will serious decline in low contrast images, making it quite difficult to locate the focus region. To tackle this challenge in computer vision, we perform autofocus by conducting a salient object detection method. Based on the mechanism of human visual system, salient object is detected by calculating global saliencies in superpixels. First, the global differences between two superpixels are computed. Then, the resulting map is optimized by introducing an inter-superpixel similarity approach. The salient object can be well detected in low contrast images. Experiments executed on three public available datasets and a nighttime image dataset prove that our model outperforms the existing state-of-the-art saliency models and has a superior performance in autofocusing application.