英语翻译First,the SIFT descriptors were clustered to form a visual dictionary .Next,each SIFT descriptor in an image was mapped to the nearest dictionary entry,and those indices were accumulated into a histogram that represented the image in term

英语翻译First,the SIFT descriptors were clustered to form a visual dictionary .Next,each SIFT descriptor in an image was mapped to the nearest dictionary entry,and those indices were accumulated into a histogram that represented the image in term
英语翻译
First,the SIFT descriptors were clustered to form a visual dictionary .Next,each SIFT descriptor in an image was mapped to the nearest dictionary entry,and those indices were accumulated into a histogram that represented the image in terms of the number of interest regions belonging to each dictionary entry.Last,a Naive Bayes probabilistic model was fitted to these histograms.This model could be applied to a new image to estimate the probability that the image belongs to any of the 9 stonefly taxa.New specimens were rejected as being non-stoneflies based on a threshold or operating point (OP) for this probability.For the results reported in this paper,we selected the OP at the equal error rate (EER; the EER OP is where the proportion of distracters rejected equals the proportion of stoneflies accepted).We estimated the OP threshold using only the training set (without distractors) by averaging over the 9 EER thresholds obtained by treating 1 stonefly taxon at a time as a distractor and using the Naive Bayes model trained on the remaining 8 taxa.
Given a new image,the system operated as follows.First,the Naive Bayes model was evaluated,and the output probability was compared to the OP threshold.If it was less than the threshold,the image was rejected as being a non-stonefly.Otherwise,the SIFT descriptors were passed to the classifier,which assigned the image to one ofthe 9 stonefly taxa.

英语翻译First,the SIFT descriptors were clustered to form a visual dictionary .Next,each SIFT descriptor in an image was mapped to the nearest dictionary entry,and those indices were accumulated into a histogram that represented the image in term
首先,SIFT描述符进行聚类形成视觉词典.其次,每个SIFT描述子在图像被映射到最近的字典中的条目,这些指数累计为在感兴趣的区域属于每个字典项的数目方面的图像直方图.最后,一个朴素贝叶斯概率模型拟合这些直方图.该模型可以应用到一个新的图像来估计图像属于任何9 Stonefly类群的概率.新的标本被拒绝作为非石蝇基于阈值或操作点（OP）这个概率.本文报道的结果,我们选择了OP的相等错误率（EER；能效比OP在哪里,拒绝的比例等于石蝇接受比例）.我们估计的OP阈值只使用训练集（无干扰）平均超过9能效比的阈值,通过一次作为分心,使用朴素贝叶斯模型的训练,剩余的8个类群1石蝇分类得到治疗.给出了一种新的图像,系统操作如下.首先,朴素贝叶斯模型进行评价,并输出概率进行比较运算的阈值.如果是小于阈值,图像被拒绝作为一个非石蝇.否则,SIFT描述符来分类,并指定图像之一9 Stonefly类群.