In this paper, we first introduce the concept of thresholded image with constant total gray level, then based on the average gradient of this new kind thresholded image, we present a new image segmentation approach including binary and multithreshold segmentation. The basic idea of this approach is that the higher the mentioned average gradient is, the higher contrast between object and background is, thus the thresholds can be determined by maximizing the average gradient. The proposed approach is more consistent with the human vision perception compared to conventional approaches based on gray-level histogram. Experimental results show the effectiveness of this approach.
Liu Hu, Liu Hu,
Mei Xie, Mei Xie,
"A new image segmentation approach based on gradient", Proc. SPIE 5985, International Conference on Space Information Technology, 598536 (4 January 2006); doi: 10.1117/12.657937; https://doi.org/10.1117/12.657937