The Thresholding Methods Based on Two-Dimensional Non-extensive Entropy

Image segmentation is one of the most critical tasks in image processing. Entropy-based threshold value is one of the most efficient techniques for image segmentation. The non-extensive (or non-additive) entropy is a recent development in statistical mechanics. In this paper, a two-dimensional Tsallis entropy (TE) with non-additive information content based on co-occurrence matrix constructed by the pairs of pixel gray value and the average gray value for the neighborhood of each pixel is applied as a general entropy formalism for information theory in image segmentation. The method based on the minimum difference of two-dimensional TE is proposed. The advantage of using average gray value for the neighborhood of each pixel as threshold value instead of gray value is discussed. It is the first time that image threshold by two-dimensional non-additive entropy is proposed to segment images. Some typical results are presented to illustrate the effect of the proposed method in the threshold segmentation.