Saliency Detection Based on the Integration of Global Contrast and Superpixels

Abstract

In the field of computer vision, the detection of salient object is an important step and one of the preconditions for salient object extraction. The outcome resulting from some existing detection methods for salient object is considerably different from the Ground Truth. In view of the shortcomings of existing methods, this paper proposes a saliency detection method based on the integration of global contrast and superpixels. The salience value of each pixel is measured according to the global contrast of the pixels in the image. A histogram optimization technique is used to highlight the low-contrast pixels of the salient region in the image and omit the high-contrast pixels of the background. In order to improve the image quality of the salient image, the superpixel image segmentation based on K-Means clustering algorithm is proposed, and finally, we generate a more accurate saliency map through the integration with superpixels. The experiment is performed on the public dataset MSRA10 K. The results show that the histogram optimization can help improve the contrast of the salient pixels and generate a better saliency map by integrating with superpixels. Compared with other classical algorithms, the proposed method outperforms other methods.