Abstract: The present two-dimensional Otsu image segmentation algorithm ignores the cohesiveness of foreground and background pixels. A novel method of threshold recognition algorithm is proposed, which, by using two-dimensional histogram of the target image, counts the absolute difference within the cluster and the average total deviation between the cluster to reflect the scattered difference, and then constructs a new threshold recognition function. An improved genetic algorithm is adopted to optimize the new threshold recognition function so as to obtain the ideal threshold value automatically. Experimental results show that the two-dimensional threshold value obtained through the optimized threshold recognition function is of good segmentation efficiency, of good retaining of the object outline and of low work amount of calculation.

Key words: image segmentation, two-dimensional Otsu algorithm, threshold recognition function, minimum within-cluster difference, maximum between-cluster difference