采用改进尺度不变特征变换在多变背景下实现快速目标识别

提出一种改进的尺度不变特征变换(SIFT)算法,用于实现多变背景下的快速目标识别。首先,构建目标图像尺度空间,提取SIFT特征点并将其按大小分类,目标识别时只需比较同一类型的特征点。然后,由SIFT特征点子区域方向直方图计算得到4个新角度用于代表特征点的方向信息,并且在目标识别时根据角度信息限制特征点匹配范围,从而提高SIFT算法的运算速度。最后,计算目标图像和待识别图像之间的尺度因子,在尺度因子约束条件下进行目标特征点匹配,从而有效地保证正确匹配数量,提高目标识别的鲁棒性。实验结果表明:当目标在待识别图像中发生局部遮挡、旋转、尺度变化或者弱光照等情况下,改进的SIFT算法能够完成多变背景下快速目标识别任务,平均识别速度提升了40%。

Fast object recognition under multiple varying background using improved SIFT method

An improved Scale Invariant Feature Transform (SIFT) method was proposed to implement the fast object recognition under a multiple varying background. Firstly, the scale space of object image was established, SIFT feature points were extracted and classified by their sizes. Only by comparing the same kinds of feature points, the target recognition could be completed. Then, four new angles were computed from the sub-region orientation histogram to represent the orientation information of each SIFT feature. Meanwhile, the feature point matching range was limited according to angle information in the target recognition to improve the calculation speeds of the SIFT algorithm. Finally, the scale factor between object image and target image was calculated and the object feature points were matched under the constraint by the scale factor to increase the number of correct matches and to insure the robustness of object recognition. Object recognition experiments were operated under object external occlusions, object rotation, scale change and illumination conditions. Results show that improved SIFT method has better performance of object recognition, and its computation speed has raised more than 40% as comparing with that of original SIFT algorithm.