CAJ | 학술논문

为了提高图像显著性检测的准确性，从数学模型上探索显著性的多特征空间。利用多尺度特征提取算法获得低层视觉特征，对特征矩阵用低秩矩阵恢复理论提取显著图，并在自底向上模型基础上融合了高层视觉特征，由高层视觉特征构成一幅权重的显著图。提高了显著度和显著目标的检测性能。通过自适应阈值算法对视觉显著目标进行分割。实验结果表明，该模型比传统的模型提取的显著目标更完整、更准确。
위료제고도상현저성검측적준학성，종수학모형상탐색현저성적다특정공간。이용다척도특정제취산법획득저층시각특정，대특정구진용저질구진회복이론제취현저도，병재자저향상모형기출상융합료고층시각특정，유고층시각특정구성일폭권중적현저도。제고료현저도화현저목표적검측성능。통과자괄응역치산법대시각현저목표진행분할。실험결과표명，해모형비전통적모형제취적현저목표경완정、경준학。
In order to improve the accuracy of image saliency detection,we explore the con-sistency of multi-features space from the view of mathematical model. We use the multi-scale feature extraction algorithm to obtain the low level visual features,introduce the theo-ry of low rank matrix recovery into the saliency map extraction,and incorporate the low lev-el visual features and the high-level visual features. The high-level visual features are fused to compose a prior map and are treated as a prior term in the objective function to improve the performance. The image saliency objects are segmented by using the adaptive threshold algorithm. Extensive experiments show that our model can comfortably achieve more per-formance to the existing methods.