融合深度及边界信息的图像目标识别

摘要

为精确定位候选目标,提高目标识别效果,提出一种融合图像边界信息和深度信息的目标识别方法,该方法可以产生数量更少、定位更准确的图像候选目标.然后提取深度学习特征,通过支持向量机分类模型,实现目标识别.在两个常用数据集上进行对比实验显示,与Baseline和选择性搜索等方法相比,该方法显著地提高了目标识别的性能.%In order to locate the candidate object accurately and improve the target recognition effect, an object recognition method combined with depth and boundary information is proposed.The proposed method can generate less but better object candidates with more accurate location.Then the depth learning feature is extracted, and the SVM classification model is used to realize the target recognition.Experimental results on two common data sets show that compared with Baseline and selective search, this method improves the performance of object recognition significantly.