目的 由于自然图像容易受到光照等因素的影响,其分割精度往往达不到人类视觉感知的需求,为此提出了一种新的结合格式塔完形规则的自然图像分割方法。方法 首先采用Ncut算法对原图像进行过分割得到若干个子区域,这些局部子区域能弱化光照、背景模糊等自然因素的影响;然后引入格式塔完形规则对区域进行度量,提出了基于区域的量化计算模型,进一步弱化了自然因素的影响,而且所得的区域率更加符合人的视觉感知;最后在区域率的基础上提出了新的合并算法,该算法简单且执行效率高,通过区域合并得到最终的分割结果。结果 30幅图像的定量和目视对比实验表明,本文算法不仅能够很好地将格式塔完形规则应用到图像分割上来,而且对比实验表明,本文算法在评价指数PRI、VOI、GCE上总体性能要优于其他算法,与人工标注的结果比较接近。结论 提出了一种结合格式塔完形规则的自然图像分割方法,该方法在过分割的基础上,采用格式塔完形规则对区域进行度量,有效降低了自然图像易受自然因素的影响,提高了分割精度。实验结果表明,本文提出的结合格式塔完形规则的图像分割算法高效性而准确,但不适合于尖细物体的自然图像的分割。

Objective In this paper, a new natural image segmentation method based on Gestalt rules is proposed to reduce the effect of illumination and other factors, as well as improve segmentation accuracy. Method The novelty of the proposed method lies in three aspects: first, the original image is segmented into several subregions with the use of normalized cut algorithm to reduce the effects of illumination, background smooth, and other natural factors. Second, the Gestalt rules are introduced to measure the regions, and a quantitative calculation model based on the regions, which can obtain the ratio region and agrees with human visual perception, is proposed. Third, a new merge algorithm based on the ratio region is proposed. The final segmentation result is obtained by merging the regions through the simple and highly efficient merge algorithm. Result Quantitative and visual inspections of 30 images show the effectiveness of the Gestalt rules on image segmentation. The results of the comparative experiments show that the effect of the algorithm is more aligned with human visual perception and that our algorithm performs better than the comparative experiments overall in the evaluation index of probabilistic Rand index, variation of information, and global consistency error. The results of our algorithm are closer to the results of artificial labels. Conclusion A natural image segmentation method based on Gestalt rules is proposed. The method adopts oversegmentation and measures the region by applying the Gestalt rules that can effectively reduce the effect of natural factors in the segmentation process and improve segmentation accuracy. Experimental results show that the proposed image segmentation algorithm based on Gestalt rules exhibits better efficiency and accuracy than other algorithms, but the effect is not good for natural image with some sharp and thin object.