中文摘要：

通过分析随机子窗口显著性检测方法的不足,提出一种改进的显著性检测方法.该方法在计算每个随机子窗口内各像素的显著性值时记录每个像素的采样次数,这样可以避免图像中间的非显著像素因多次采样而导致其显著性值过高;通过计算一次每个像素与整幅图像均值之间的距离,从而保证每个像素都参与显著性计算;最后运用积分图像加速每个子窗口内的特征均值计算.实验结果表明：相比于传统的随机子窗口显著性检测方法,所提方法的检测效果更好,而且计算时间也显著下降.

英文摘要：

An improved saliency detection method was proposed after traditional random sub-windows saliency detection method was analyzed.In this method,the sampling number of all pixels were recorded when the salience value of each pixel was computed in each random sub-window to avoid that the salience values of the non-salient pixels in the middle of the image were too high due to multiple sampling.It was ensured that all pixels were participated in the saliency calculation by computing the distance between each pixel and the mean of the entire image.Finally,the integral image was used to accelerate the computation of the mean in each random sub-window.The experimental results show that compared with traditional random sub-window saliency detection method,the proposed method achieves better detection effect and the computation time is also significantly decreased.