基于多特征和流形排序的显著性检测

显著性检测近年来得到广泛关注,其中基于流形排序的显著性检测取得了较为理想的效果。但该方法用颜色平均值来描述图像单元,这就使得图像单元内的细节信息被忽略。此外,根据人类视觉系统,人眼对颜色、纹理等多种特征敏感,而单一的颜色均值信息只能从单一角度描述图像,不能够全面立体地刻画图像内容。因此,提出了采用多个特征从多个角度对图像内容进行刻画。在公共数据集上的实验结果表明,所提出方法能够得到较好的检测效果,优于所有对比方法。

Saliency Detection Based on Multiple Features and Manifold Ranking

Saliency detection has been paid more and more attention in recent years, and manifold ranking based saliency detection has achieved satisfactory results. However, this method used the mean value in Lab color space to describe the image unit, which makes the detail information be ignored. In addition, according to the human visual system, the human eyes are sensitive to color and texture, and the mean value can only describe the image from a single view, but cannot describe the image content adequately. Therefore, in this paper, multiple features from multiple views are used to describe the image content. The experimental results on the public dataset show that the proposed method can obtain better detection results, which is superior to all other compared methods.