一种新型的不变性快速图像识别方法

采用平面直角坐标系和其相对转化型式极坐标系 ,利用极角和极值的二重三阶累积量构造出二维的三阶累积量特征集合 ,并对该特征集进行快速 FFT变换得到频域的特征变换集合 .该特征集具有数据量少 ,保留了每幅图像的内部相位信息 ,又对图像的平移、旋转、比例和噪声变化不敏感的特点 .利用神经网络 ,实现了图像快速、并行、准确的识别 ,并有利于实现大量图像样本的存储 .实验中每副图像在提取30个特征的情况下 ,图像可较好地得以识别

A new instant recognition method of invariable images

Employing double third order cumulants of polar angles and extremes in a planar rectangular coordinate system and a polar one relevantly conversed from the former, a two dimensional characteristic aggregate (CA) of third order cumulants is constructed and, then, a characteristic mapped aggregate of frequency field is obtained by conducting fast FFT transformation upon the CA. This aggregate is featured with little data amount, longer time of keeping internal phase messages of each set of image, and low sensitivity to translation, rotation, scaling, and noise of the picture. Neural network has been used to realize fast, parallel, and accurate identification of the image and it is useful for realization of large storage of image patterns. The experiment shows that the image can be well identified while 30 characters are picked up from each image.