Feature representation of natural images based on contours

An approach of image feature representation that combines the multi-scale image decomposition technique and contour theory is proposed. Its advantage lies in the fact that the empirical mode decomposition technique is applied to the gradient magnitude of the image, not to the image directly. The gradient magnitude image is decomposed by the empirical mode decomposition technique to obtain a number of intrinsic mode functions that capture the gray level change information under different scale of the image. In order to represent the features of the image perfectly, the first two intrinsic mode functions are summed and the results are described by contours. The experiment results show that the proposed representation method for image features not only captures the gray level change information under different scale of image, but also catches the geometric structure information of image, and the weak features of an image can be expressed better.