提出一种基于局部特征提取的目标识别方法, 用于自动识别不同尺度, 视角和照度条件下的目标。首先,建立图像的尺度空间; 结合海森矩阵和Harris算法提取局部特征点, 计算该特征区域的方向和灰度梯度及方向; 统计出每块子区域的标准灰度梯度直方图, 得到128维的特征向量。然后, 基于主成分分析的降维算法来降低特征向量的维数, 加快识别的计算速度。最后, 采用特征空间分类器增加目标识别的速度。实验结果表明：基于局部特征提取的目标识别达到了较高的识别率, 在视角、尺度和照度变化下的识别率分别为61.9%, 80.5%和84.4%, 平均识别时间为130.9 ms。与尺度不变特征变换(SIFT)和加速鲁棒特征(SURF)算法相比, 本算法不仅在不同的视角, 目标尺度及照度条件下具有较高识别率, 而且识别速度比SIFT方法高。

A target recognition method was proposed to recognize targets with different scales, view-points and illuminations automatically. First, a scale space of images was established, and the local key points in the scale space were extracted by incorporating the Hessian and Harris scale-space detectors. Then, the main orientations of the key points and orientation histograms were calculated and 128 element feature vectors for each key point were established, in which these feature vectors were invariant in different rotations and illuminants. To reinforce the performance, principle component analysis was incorporated to reduce the dimensionality of feature vectors and improve calculating speeds for the recognition. The nearest feature space classifier was used for increasing the recognition speeds in robustness. Experiment results show that this proposed method achieves a significant improvement in automatic target recognition rate, and the recognition rates for varied view-points, scales and illuminations are 61.9%, 80.5%, and 84.4%, respectively. Compared with the Scale Invariant Feature Transform(SIFT) and Speeded Up Robust Features (SURF), the proposed method achieves a significant improvement in automatic target recognition rate in presence of varying viewpoints, scales and illuminations.