选择性背景优先的显著性检测模型

Saliency Detected Model Based on Selective Edges Prior

在检测图像显著性区域的领域中,背景优先是一个较新的思路,但会遇到背景鉴别这个具有挑战性的难题。该文提出背景真实性的判断问题,在探索的过程中发现背景通常具有连续性的特征,根据这一特性实现了判定背景的方法,并将判断的结果作为显著性先验值应用于后继的计算中,最终结果的准确性和正确性得到有效提高。该文首先采用均值漂移(MS)分割算法将图片预分为超像素,计算所有超像素的初始显著值;随后提取原图的4个边界条,计算每两条之间的色彩直方图距离,判定小于预设阈值的两条边界作为"真"的背景,选择它们作为优先边界,计算先验显著性值;最后进行显著性计算,得到最终的显著图。实验结果表明,该算法能够准确检测出显著性区域,与其他6种算法相比具有较大优势。

In the field of saliency detection, background prior has become a novel viewpoint, but how to identify the real background is challenging. In this paper, a background-identified method is proposed based on homology continuity using the extracted background features, and the identified background is applied to the following computation, improving the eventual saliency map in accuracy as well as correctness. First, the primary saliency of each superpixel produced by Mean Shift（MS） segmentation algorithm is calculated. Second, 4 edges are extracted to generate their RGB histograms, and the Euclidean distance between each two of the histograms is calculated, if the distance is smaller than a given value, these two edges are defined to be continual and more likely to be the real background. Finally, the pixel’s saliency is calculated using the prior background knowledge to figure the final saliency map. The results show that the proposed method outperforms other algorithms in accuracy and efficiency.