Abstract: The research work of image processing and recognition by means of non-linear chaotic
method is receiving increasing attention. In the existing literature, there has been a method which
constructs dynamic system by taking sinusoidal function as auxiliary function and image, and
iteratively generates chaotic attractors as image features. In order to further explore the characteristics
of image attractors as image features and improve the recognition effect, this paper uses a discrete
cosine transform (DCT) basis function matrix instead of a sine function to generate approximate
chaotic attractors iteratively for face recognition. First, this study analyzes the diversity and
oscillation of DCT basis function matrix. Then, the DCT basis function matrix and the image matrix
are used to construct the iterative expression, and the proposed iterative algorithm is used to generate
the attractor. After the attractor is transformed by fast Fourier transform, the correlation coefficient is
calculated, and the face image is recognized. For the Yalefaces image database, when each image can
be trained, the recognition rate can reach 100%. When the first five images of each group are trained
to extract the feature, the recognition rate can exceed 85%. For CMU PIE databases, when each image
can be trained, the recognition rate can exceed 99%. And this attractor method can be used as a
method of image bottom feature extraction, which still needs further study.

Key words: face recognition, discrete cosine transform basis function, chaotic attractor, image features