Workpiece Invariant Pattern Recognition Based On Orthogonal Fourier-mellin Moments

For the vision of assembly robot, realizing the invariant shape recognition of mechanical workpiece is the base. Doing research in invariant recognition can improve capability and adaptability to the environment for robot, reduce the recognition error and largely improve the successful rate of the assembly. During the system design, this paper uses the way of orthogonal Fourier-Mellin moments to extract the image feature, and this conducts the reliable recognition of mechanical workpiece.Firstly, we develop video capture program completely by software method with VFW software package provided by Microsoft Corporation, and in this way the generality of the program is improved.Secondly, in order to eliminate image noise, enhance the definition and separate the object from background, a series of pre-processings are carried out to the images captured, such as image smoothing, median filter, grad sharp, threshold segmenting edge tracking and seed filling. This makes preparations for the next step of feature extraction.Thirdly, computering the OFFM's and normalizing the algorithm are conducted to the images pre-processed. The detailed algorithm is shown as follows:first we determinate the center of the image by using the fist-order geometrical moment. The center of the image is used as the origin of the coordinate system; then we calculate the FFM's in the Cartesian coordinate system, using the corresponding complex moments. And we normalize the FFM's against changes in scale and intensity; last we get the normalized OFFM's by the normalized FFM's, and calculate the modulus of the OFFM as the image features. This type of features has the advantage of being invariant to dimension, orientation and intensity variance.At last, we employ the way of Euclidean distance to classify the images. The particular method showed as: first capturing a series of images to extract the image feature and train the classifier; then test the recognition result by capturing another group of images.Experimental results have shown the effectiveness of this system.