摘要
为了有效地对步态特征进行分类识别,提出了一种基于Radon变换和解析Fourier-Mellin变换的步态识别算法。该算法直接对在视频序列中检测到的灰度图像进行Radon变换,然后进一步对变换结果进行解析Fourier-Mellin变换,从而将原图像的旋转变化和尺度变化分别转化为相位变化和幅度变化,之后通过定义一族旋转与尺度不变函数提取目标图像的不变性特征进行分类识别。实验结果表明,与目前常用的基于Hu矩和Zernike矩的算法相比,由于不需要对目标图像二值化和归一化,从而可以保留图像的更多细节信息,避免了重采样与重量化误差,该算法应用于步态识别有更高的识别率,可以达到更好的识别效果。
In order to have an effective classification and recognition on the gait characteristics, a gait recognition algorithm based on radon and Fourier-Mellin transforms is proposed. The method can directly utilize the Radon transform to project the image onto projection space. Then the analytic Fourier-Mellin transform is used to the projection space to convert the rotation ;~nd scaling of the original image to a phase shift and a scaling of ampli- tude. After then, a rotation and scaling invariant function is constructed to extract invariant features from target images. At last, a classifier is employed to classify the features. The experimental results show that compared with the Hu moment and Zernike moment methods, the proposed method doesn＇ t need binarization and normalization, and it can avoid re-sampling and re-quantization, reserve more detail information, so the method is applied to the gait recognition to have a higher recognition rate, and attain a more better effect.
出处
《电视技术》
北大核心
2014年第5期169-172,208,共5页
Video Engineering
基金
国家自然科学基金项目(61162020)
宁夏自然科学基金项目(nz1138)
