Abstract: For the significant detection algorithms that existed at this stage, the contrast is not clear, the image details of the significant images are not obvious enough, and the background suppression is not complete, this paper proposes a full-frequency saliency detection CRT algorithm based on contrast stretching algorithm and Laplace algorithm to improve Robert cross gradient sharpening. This method uses FT algorithm to calculate its own fast speed and has the advantage of de-drying effect. Contrast stretch enhances the contrast of the significant image and the features of the Robert gradient sharpening algorithm to improve the edge detail and the background correction algorithm proposed in this paper can further suppress the background. It finally gets better saliency map with higher contrast, more obvious detail, and background suppression. The algorithm uses the MSRA10K and THUR15K data sets for significant detection testing, compared with the other six popular saliency detection methods. From the subjective point of view, the contrast of the significant image is obtained, and the image details are superior to the six algorithms compared with it. The objective indicators show that the MAE value obtained by the MSRA10K data set for the significance detection is 0.12, and the MAE value obtained on the THUR15K data set is 0.06, which is better than the six algorithms compared with it. The average structural index S-measure value is 0.8, which is only slightly lower than the HC algorithm, which is better than the other five algorithms. That is to say, the saliency map obtained by the algorithm has the advantages of contrast and image detail enhancement, and has better background suppression effect.

Key words: significant detection algorithmst, full-frequency significance detection algorithm, contrast stretch algorithm, Robert cross gradient sharpening, background suppression effect

摘要： 针对现阶段已经存在的显著性检测算法存在对比度不鲜明，显著图的图像细节不够明显，背景抑制不彻底的不足，提出一种基于对比度拉伸算法和用拉普拉斯算法改进Robert交叉梯度锐化的全频域显著性检测CRT算法，该方法利用FT算法自身运算速度快，具有去噪效果的优点以及对比度拉伸提升显著图对比度以及Robert梯度锐化算法提升边缘细节的特点以及提出的背景修正算法对背景能够进一步抑制的特点，最终将获得对比度更高，显著图细节更明显，背景抑制效果更好的显著图。该算法使用MSRA10K和THUR15K数据集做显著性检测检验，与其他6种流行的显著性检测方法做对比。从主观上得到显著图的对比度，图像细节均优秀于与它对比的6种算法。客观的指标显示，该算法用MSRA10K数据集进行显著性检测得到的MAE值是0.12，在THUR15K数据集上得到的MAE值是0.06，均优于与它对比的6种算法。平均结构性指标S-measure值为0.8，只略低于HC算法，优于其他5种算法。即该算法得到的显著图既具有对比度和图像细节增强，而且具有背景抑制效果更好的优点。

关键词: 目标显著性检测, 全频域显著性检测算法, 对比度拉伸算法, Robert交叉梯度锐化, 背景修正方法