Invariant Object Recognition using Eigenvalues of Covariance Matrices and Autocorrelation

T.H. Sun and F.C. Tien (Taiwan)


2D object recognition, eigenvalues of covariance matrix; autocorrelation coefficient; feature extraction


This paper presents a new boundary-based part recognition method for two-dimensional part. The proposed method adopts the eigenvalues of covariance matrix, re-sampling and transformation of autocorrelation coefficient for feature extraction and the simple minimum Euclidean distance for object classification. The boundary of the binary digital object is represented into the form of the eigenvalues of covariance matrix over a given region of support, and then is further transformed with autocorrelation function. The derived features are unique and invariant to translation, rotation, and scaling. Finally, the minimum Euclidean distance is used for pattern recognition for simplicity. Twenty-five standard patterns are acquired, and for each object ten extra images using different positions, orientation and scales are digitized for system verification. The experimental results show that the proposed system achieves the high recognition rate.

Important Links:

Go Back