基于频率域的图像显著性区域提取方法

Based on the Frequency Domain Image Salient Region Extraction Method

针对图像的显著性区域提取方法存在区域边界不明确、分辨率不高、显著性目标不准确等问题,提出了一种使用高斯差分滤波器的组合来保持原图像更多的频率信息,然后在频率域里利用颜色和亮度两个特征来计算显著图,最后用均值偏移方法结合自适应阈值方法从显著图中提取显著性区域。实验结果表明,使用高斯滤波器的方法不仅输出完整分辨率的显著图,而且能准确地突出显著性目标,使区域具有明确的边界,并且分割的查准率和查全率都好于多尺度分析方法和Itti方法。

An image salient region extraction method based on the frequency domain was proposed. This method used a combination of DoG filters to retain more frequency contents from the original images, and used color and brightness information to estimate the salient region. We used the mean shift method combined with adaptive thresh- old method from the saliency map extract salient region at last. The experimental results show that this method not on- ly outputs full -resolution salient map, but also solves the consistently highlight salient region, high resolution and gives well - defined boundaries of salient objects. The method performed that on the segmentation task by achieving both higher precision and better recall than multiscale analysis method and Itti＇s method.