基于视觉注意的SVM彩色图像分割方法

SVM model for segmentation of color image based on visual attention

提出一种基于视觉注意的自然场景彩色图像支持向量机(Support Vector Machine,SVM)分割方法。基于人类视觉注意机制将图像进行预分割,得到图像的显著区域和非显著区域,利用形态学操作对得到的图像进行处理,并自动选取和标注SVM的训练样本,用训练后的SVM分类器对整幅图像进行分割。该方法充分利用视觉注意机制方法的有效信息,解决了其边界不确定的缺陷,并且结合具有很好泛化性能的SVM学习方法,在无需先验知识以及任何人工干预的情况下,实现对自然场景图像的分割。为验证算法的有效性,分别从加州大学伯克利分校图像数据库及互联网选取多幅彩色图像进行实验,实验结果表明:该方法的分割结果不仅与人类视觉注意结果相一致,而且与伯克利图像数据库中人工标注结果相比,得到较好分割效果。

An SVM（Support Vector Machine） model for the segmentation of color images of natural scenes based on visual attention is proposed.Based on human visual attention mechanism,the image is pre-divided,and saliency and non-saliency areas are obtained.The pre-divided image is further processed by some operators of morphology such that the training samples for SVM can be chosen and labeled automatically.The whole image is divided by SVM classifier.The proposed approach can solve the problem of uncertain boundary through effective information of the image based on visual attention mechanism. Combined with SVM,the approach can cut the image well without any priori knowledge and manual intervention.To evaluate the performance of the presented approach,some experiments on several images from the image dataset of University of California at Berkeley and Internet are accomplished.The experimental results demonstrate that not only the segmentations are consistent with the habits of the human visual attention,but also the proposed approach can obtain good segmentation results comparing with that of manual labeled.