基于显著性检测的图像单分类算法

One-class Classification Algorithm Based on Saliency Detection

针对目前图像单分类的算法受复杂背景的干扰较大,背景中存在与待分类物体相似的特征,结合当前先进的显著性检测技术与图像单分类算法,提出了基于显著性检测的图像单分类算法。利用现在先进的金字塔特征提取网络,通过注意力机制对图像进行显著性检测,得到显著区域的掩膜图。将掩膜图与原图进行与操作,获得待分类目标。将处理得到的图片送入分类网络,利用ALOCC算法实现图像分类。实验表明,该方法在图像单分类上具有较好的效果,分类准确率得到提高。

Aiming at the current one-class classification algorithm is greatly interfered by the complex background,the characteristics of image have high similarity with background,combining with the current advanced saliency detection technology and the one-class classification algorithm,proposing a one-class classification algorithm based on saliency detection in this paper.Take advanced pyramid feature extraction network,obtain the mask image of the salient area by using the attention mechanism.combined the mask image and the original image to obtain the target.Send the processed pictures to the classification network,and use the ALOCC algorithm to achieve image classification.The experiments show that this method has a good effect on one-class classification and improves classification accuracy.