隐马尔可夫模型下基于SIFT特征的局部遮挡目标识别

Partially Occluded Object Recognition Based on SIFT Features under Hidden Markov Model

提出一种利用隐马尔可夫模型建立目标特征匹配库来识别图像中局部遮挡目标的新方法。该方法首先通过SIFT算法提取目标SIFT特征,然后采用隐马尔可夫模型对目标所有的SIFT特征进行训练,得到目标SIFT特征对应的模型输出概率范围,将该概率范围作为目标特征匹配库。在对图像中的目标进行识别时,利用目标特征匹配库可以把目标特征从图像所有特征中识别出来,即使目标遮挡比例为60%时,该方法仍能识别出目标。实验结果表明,新方法可以精准地识别出图像中被遮挡目标,能够很好地解决遮挡情况下的目标识别问题。与现有局部遮挡目标识别算法相比,新方法所取得的目标识别率均有所提高。

This paper proposed a new method to recognize partially occluded image object via feature matching library by using Hidden Markov Model（HMM）. Firstly, the object SIFT features extracted by SIFT algorithm were trained by HMM, and the probability range corresponding to the SIFT features was obtained as the object features matching library. The matching library can identify object SIFT features from all image SIFT features during image object recognition, even when the object occlusion ratio reaches 60%, the new method can still identify the object. The experimental results indicate that the new method can accurately identify partially occluded image object, and it can be a good solution in case of partially occluded object recognition problem. Compared with the existing partial occlusion object recognition algorithm, the new method can improve object recognition rate.