【摘要】 提出一种新的基于纹理特征相似度的图像区域分割方法——迭代合并-扩张法。通过对各区域直方图的相似性比较,来确定不同区域纹理特征的相似性,把纹理特征相似性权值作为区域一致性度量,迭代合并相似区域完成图像的初始分割,再通过区域扩张来解决小区域纹理特征的不稳定性问题,完成图像中水域与陆地的精确分割。最后利用形态学对比度法在水域中进行目标检测。试验结果表明,此方法能很好地实现复杂海地背景下的舰船目标检测。更多还原

【Abstract】 An iterated merging-and-expanding algorithm of image region segmentation based on texture similarities was proposed. The texture similarity of different regions was evaluated by comparing their histograms. The weights of texture feature similarities were treated as the measurement of region consistency, and the initial region segmentation was accomplished by iterated merging similarity regions. Then, the process of expansion was done to avoid the instability of texture features in smaller regions and the precision segmentations of water-area and land were achieved. Finally, the operation of target detection in the water-area was performed based on morphologic-contrast. The results show that the ship targets in complicated background of sea and land can be effectively detected by using this algorithm.更多还原