Image representation in differential space

Abstract

In this paper, a derivative estimator is introduced to obtain differential information of images. Experiments show that differentials obtained by this estimator outperform the traditional Sobel operator and this estimator is practical for extracting differential image information. A new image representation in this differential space is also proposed. Differential sign sequences of images are used as the signature of image patterns. The Hamming distance is used for template matching. The proposed representation is invariant to brightness and contrast and is robust to noise because of the low pass property of the estimator. Template matching is used as an example to exhibit the advantage of this representation. Experiments demonstrate good performance of the proposed method. © 2008 Springer Berlin Heidelberg.