【摘要】 提出一种基于类物体区域检测的BoW(Bag of Words)框架物体识别方法,采用改进的BING(Binarized Normed Gradients)算子检测分割出图像中的可能物体区域后,利用RootSIFT算子提取特征,送入后续BoW框架进行物体类别识别.将该方法应用于PASCAL VOC2007图像集,试验结果表明:相较于整幅图像的特征计算,将特征提取与匹配限定在固定的可能物体区域的做法可以提高计算速度和识别效率.此外,该方法在VOC2007图像集上达到了平均33.45%的识别准确率,优于相关文献算法.更多还原

【Abstract】 In this study,an object recognition method for bag of words(BoW)framework is proposed via objectness measurement.Firstly,the object areas are detected and segmented using the improved binarized normed gradient(BING)operator.Then,the features are extracted by RootSIFT operator for object recognition.Finally,this method is employed for PASCAL VOC2007 image-set.Therefore,it is found from experimental results that,compared with the whole image feature computation,the computational speed and recognition efficiency are enhanced by feature extraction and matching limitation to possible object areas.In addition,the algorithm in this approach is proven better than those in other literatures with average recognition accuracy of 33.45% for VOC2007 image-set.更多还原