Abstract: The visual prominent image refers to the image which has a salient object. Compared with those images that have messy contents, such images usually attract users’ more attention and conform to more needs of users when they retrieve images. A method which is used to classify image object visual saliency is proposed. It builds an object visual saliency classifier model by support vector machine on the basis of “center-ground” computational principle and multi-feature combination. Also, the score describing the degree of saliency of the image can be given by this model. Experimental results demonstrate that the model has higher classifier accuracy and is closer to manual operation results on web image applications, such as resorting in image retrieval, automatic audit images uploaded.

Key words: visual object saliency, “center-ground&rdquo, operation, Local Binary Pattern（LBP）, visual saliency map, Support Vector Machine（SVM）

摘要： 从视觉角度来说，视觉显著性图像是指主体突出的图像，比起内容散乱的图像，此类图像往往更能吸引用户的关注，也更符合用户对图片检索的使用需求。提出了一种图像主体视觉显著性判断方法，采用“中心围绕”计算原则在多特征融合的基础上应用支持向量机训练，建立了一个分类模型，并且可以给出表征图像显著程度的得分。实验表明，该模型有较高的分类正确率，并且将该模型应用于图像检索重排序、图像上传自动审核等应用时，可以得到更接近人工操作的结果，降低人力资源成本。

关键词: 视觉主体显著性, &ldquo, 中心围绕&rdquo, 操作, 局部二值模式, 视觉显著图, 支持向量机