基于物体信息的图像显著性区域检测

针对目前图像显著性检测存在的显著性图边缘模糊、缺乏视觉高层信息等问题,提出了一种基于物体信息的显著性检测算法。将似物性计算与显著性计算相结合,可以将前景与背景分离,抑制高对比度干扰区域的影响。首先采用改进的L0平滑算法对图像进行滤波处理,然后通过超像素分割将图像分成若干图像块,再通过聚类算法进行合并,得到待检测图像块。采用似物性检测算法计算物体可能存在的区域,与待检测图像块进行融合,得到物体显著性图,再通过颜色对比度及空间分布特征计算显著性图,最后将二者融合,得到最终结果。实验结果表明,算法能够得到清晰的物体边缘,对图像的显著性区域能够较为全面地覆盖,有效地抑制高对比区域的干扰。

Salient Region Detection Based on Objectness Cue

An algorithm for salient region detection based on objectness cue is proposed for blur of edges in image and lack of high level information in recent research of saliency detection. Objectness and saliency are combined for separating foreground from background and restrain noise from high contrast regions in image. Firstly, a modified L0 norm smoothing algorithm is adopted for image filtering. Secondly, superpixels segmentation is executed obtaining amount of image regions. Thirdly, superpixels regions are merged through clustering algorithm. Objectness detection algorithm is applied for searching generic object proposals and combining with image regions for objectness saliency map. Saliency of low level features are computed through color contrast and spatial distribution. Finally, objectness saliency and low level feature saliency are combined for saliency map. The experiments show that the proposed algorithm can get saliency map with clear edges, cover all salient regions in image and restrain noise from high contrast regions effectively.